Research ArticleHybridized Extreme Learning Machine Model with Salp SwarmAlgorithmANovel PredictiveModel forHydrological Application

Zaher Mundher Yaseen 1 Hossam Faris2 and Nadhir Al-Ansari 3

1Sustainable Developments in Civil Engineering Research Group Faculty of Civil Engineering Ton Duc ang UniversityHo Chi Minh City Vietnam2King Abdullah II School for Information Technology e University of Jordan Amman Jordan3Civil Environmental and Natural Resources Engineering Lulea University of Technology 97187 Lulea Sweden

Correspondence should be addressed to Zaher Mundher Yaseen yaseentdtueduvn

Received 8 August 2019 Accepted 16 January 2020 Published 21 February 2020

Guest Editor Jesus Vega

Copyright copy 2020 ZaherMundher Yaseen et al(is is an open access article distributed under the Creative CommonsAttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

(e capability of the extreme learning machine (ELM) model in modeling stochastic nonlinear and complex hydrologicalengineering problems has been proven remarkably (e classical ELM training algorithm is based on a nontuned and randomprocedure that might not be efficient in convergence of excellent performance or possible entrapment in the local minimaproblem (is current study investigates the integration of a newly explored metaheuristic algorithm (ie Salp Swarm Algorithm(SSA)) with the ELM model to forecast monthly river flow Twenty years of river flow data time series of the Tigris river at theBaghdad station Iraq is used as a case study Different input combinations are applied for constructing the predictive modelsbased on antecedent values (e results are evaluated based on several statistical measures and graphical presentations (e riverflow forecast accuracy of SSA-ELM outperformed the classical ELM and other artificial intelligence (AI) models Over the testingphase the proposed SSA-ELM model yielded a satisfactory enhancement in the level accuracies (84 and 131 percentage ofaugmentation for RMSE and MAE respectively) against the classical ELMmodel In summary the study ascertains that the SSA-ELM model is a qualified data-intelligent model for monthly river flow prediction at the Tigris river Iraq

1 Introduction

Due to the wide range of space and time variabilities riverflow modeling has been a critical problem in water man-agement and hydrology [1] (e operation of many waterresources system rule curves needs monthly river flowforecasts [2] Various methods capable of forecasting riverflow under different conditions with different levels of ac-curacy have been proposed considerably evolving within thepast decades from the simple linear equations to the complexand complicated ones Researchers have considered variousriver flow forecasting methods such as the conceptual (HEC-HMS and HBV) [3ndash5] stochastic (AR-autoregressive andARMA-autoregressive moving average) [6] and the phys-ical-based (SWAT) models [7 8]

(e complex relationship between rainfall and river flowvariables and the lack of sufficient hydrological data ofwatersheds have made data-driven models such as artificialintelligence more suitable for river flow forecasting com-pared to the process-based models [9] Currently a singlemethod which can be employed generally for all types ofbasins and systems is yet to be developed Hydrologists havecurrently been building reliable forecasting methods byexploiting the available and new scientific techniques such ashybrid intelligence models (e ability of hybrid AI modelsin addressing the associated nonlinearity and non-stationarity problems has attracted great attention in dif-ferent domains [10 11] Likewise this explains their wideapplications in modeling the river flow in particular [12 13]and hydrological processes in general [14]

HindawiComplexityVolume 2020 Article ID 8206245 14 pageshttpsdoiorg10115520208206245

Studies have demonstrated a wide range of forecastingaccuracies using AI models with the geographic topogra-phies of the test sites despite their success in other fields[15ndash18] (e development of a model that will be used toforecast in any environmental condition is an aspect yet to beexplored Researchers in water resources management en-gineering should be motivated by the assertion does theintegration of an optimization algorithm with AI models ashybrid intelligent models boost their predictive model ac-curacies Hydrologists and intelligence system modelers inan attempt to address the raised issue have faced thechallenge of properly selecting the input combinationsknown to rule the predictable accuracy of an objectivepredictive model for model development [19]

(e literature stated a very massive implementation ofthe AI models for river flow modeling (ese models areartificial neural network genetic programming fuzzy logicmethods support vector machine models M5 tree modelscomplementary AI models coupled with data time seriespreprocessing techniques and hybrid model [14] Mostrecently a new version of the ANN model called extremelearning machine was developed by the authors in [20] as arobust predictive model (e employment of ELM in thefield of hydrology and particularly in river flow modelingdemonstrated a very promising and enhancement in themodeling process [14]

(e first attempt of modeling river flow using ELM wasestablished in [21] (e authors developed an ELM model tocapture the associated nonlinearity of the seasonal riverinflow of the Brazilian Hydropower (eir Finding evi-denced the potential of the proposed model and emphasisedseveral hydrological investigations later on Li and Cheng[22]studied the capacity of ELM tomodule themonthly scalereservoir inflow in China (e ELM model provided anaccurate result in comparison with several other well-established AI models An improved ELM model is devel-oped in a new version called online sequential extremelearning machine (OS-ELM) for multiple river flow scalesprediction in Canada [22] (e same OS-ELM model wasdeveloped for flood events forecasting for hourly river flowmonitoring [23] (e findings of the improved ELM modeldemonstrated a noticeable prediction performance In an-other attempt Yaseen et al developed an ELM model toforecast the monthly scale river flow located in semiaridenvironment in Iraq (e proposed model revealed goodcapability to mimic the monthly river flow trend Later onseveral studies approved the potential of the ELM in the fieldof hydrology processes [13 24ndash29] Based on the reportedliterature of the extreme learning machine modeling and itsversion improvement in hydrological problems over a veryshort period the authorsrsquo attention is mainly to modify thisrobust predictive model and attempt to enhance its pre-dictability skills

Despite the noticeable promising implementation of theELM model in wide range of hydrology applications it stillsuffers limitations and drawbacks First the random as-signment of the input weights and biases can negativelyaffect the generalization capability of the network [30] Inaddition ELM needs larger number of hidden neurons

which results in more complex models [31] Hence the mainmotivation of the current research is to propose a newtraining optimization for the ELM model to solve theaforementioned problems and improve the predictabilityperformance of the ELM model

In this research a new optimization algorithm calledSalp Swarm Algorithm is proposed and hybridized with theELM model for optimizing the input weights and hiddenbiases of the network SSA is a modern nature-inspiredalgorithm proposed in [32] SSA has shown an impressiveperformance in optimizing various complex and challengingengineering problems Based on these facts the motivationof exploring the capabilities of this optimizer is endeavouredand the authors work to improve the performance of theclassical ELM model for the surface hydrology applicationFor this reason a real case of semiarid environment of theTigris river in Iraq is investigated which is an extension forthe published work [33]

(is research is organized as following A brief intro-ductory of the proposed methodology is given in Section 2(e description of the investigated case study is described inSection 3 (e proposed ELM model based on the SSAoptimizers is introduced and discussed in detail in Section 4(e experiments and results are given and discussed inSection 5 Finally the conclusions and findings of this workare summarized in Section 6

2 Preliminaries

21 Salp Swarm Algorithm SSA is a recent nature-inspiredalgorithm which mimics the swarming behaviour of seacreatures called salps [32] Salps have barrel-shaped gelat-inous bodies and they move around by pumping waterthrough their body from one side to the other Salps exist ascolonies and move together as chains In SSA the movementbehaviour is mathematically modelled for solving optimi-zation problems (e models suppose that there are twomain types of salps leaders and followers(e leaders are thesalps that lead the chain in the front direction while thefollowers follow their leaders synchronously and in harmonyas shown in Figure 1

Like all swarm intelligent algorithms SSA starts byrandomly initializing a swarm of N salps Each salp isrepresented by one-dimensional vector of n elements whilethe swarm is represented as a two-dimensional matrix x (ealgorithm also refers to the target of the swarm as a foodsource F(emovement of the leader chain x1

j is modelled asgiven in the following formula

x1j

Fj + c1 ubj minus lbj1113872 1113873c2 + lbj1113872 1113873 c3 ge 05

Fj + c1 ubj minus lbj1113872 1113873c2 + lbj1113872 1113873 c3 lt 05

⎧⎪⎨

⎪⎩(1)

where j is the dimension of the position to be updated Fj isthe jth element of the food source position ubj and lbj arethe upper and lower bounds of the jth element c2 and c3 arethe random numbers drawn from the interval [0 1] and c1 isthe dynamic variable that changes its value over the course ofiterations according to the following formula

2 Complexity

c1 2eminus (4lL)2

(2)

in which l represents the current iteration and L is a pre-defined number of maximum iterations Note that c1 is avery important variable in the SSA process which controlsthe balance between the exploration and exploitation pro-cesses of the optimization algorithm (e followers updatetheir position accordingly as

xij 05 x

ij minus x

iminus 1j1113872 1113873 (3)

where xij is the ith element of the jth position and ige 2 (e

procedure of the SSA algorithm is presented in Figure 2

22 Extreme LearningMachines Artificial neural networks(ANN) have been used as a universal approximator to dealwith complex problems in a wide spectrum of differentdomains One of these common problems is classificationand regression problems For ANN to be able to predictthe value of the output variable it needs to be trainedbased on a set of examples using a learning algorithm Oneof the recent and successful variants of ANNs is ELMELM is a learning framework for single hidden layerfeedforward neural networks (SLFN) introduced recentlyin [20] (e main merit of the ELM model is the learningprocess convergence speed that emphasizes the capabilityover the classical gradient decent approach for trainingthe SLFN [34] In addition the nonrequirement for al-gorithm tuning parameters is similar to the back-propagation algorithm (e ELM model requires nohuman interactive for tuning [35] Further ELM does notrequire an iterative process to tune the SLFN parametersInstead it randomly modifies the input weights and thehidden biases and then it governs the output weightssystematically using the potential of the MoorendashPenrose(MP) generalized inverse method An illustration of theSLFN-based ELM is shown in Figure 3

(e output of SLFN is trained using a dataset with Ndistinct examples denoted as (xi ti) i 1 2 N where xi isthe input vector and ti is the output vector (e activationfunction g() and k hidden neurons is

fi 1113944k

j1βjg wjxi + bj1113872 1113873 j 1 2 N (4)

where wj denotes the weight vector that links the inputnodes to the jth hidden node bj signifies the bias value of thejth hidden node and βj are the values of the output weightsthat connect the jth hidden node with the output nodes [36]Formula (4) is written compactly as

O H times β (5)

where H is the output matrix of the hidden layer ELMassigns the weights and the biases randomly without theneed to take into consideration the input data ELM de-termines the output weights by solving equation (5) using

Movement direction

Leader salp

Follower salps

Figure 1 Leaders and followers of salps chain movement

Initialize a swarm of slaps xi(i = 1 2 n)While (Termination condition is not met) do

Calculate the fitness of each salp in the swarmSet F as the position of best salpUpdate c1 by equation (2)for (each salp (xi)) do

if (i == 1) then

else

end if

end whileReturn F

end for

Update the leaderrsquos position using equation (1)

Update the followerrsquos position using equation (3)

Check and return salps if they go beyond upper andlower bounds

Pseudocode of the SSA algorithm

Figure 2 Salp Swarm Algorithm procedure

Complexity 3

MP generalized inverse (e output weights can be deter-mined by ELM using the following formula

1113954β Hdagger

times T (6)

(e process of the ELM predictive model is presented inFigure 4

3 Case Study and Site Description

In this study the Tigris river which is located in the Iraqregion is selected for examining the proposed hybrid pre-dictive model (e total length of the river is around1718 km (e river originates from Turkey and flows towardIraq in the southern part Most of the river basin (about 85percentage) is located in Iraq with an approximate catch-ment area of 253000 km2 Tigris and Euphrates rivers arepresenting the chief freshwater resources in Iraq providingwater for multiple usages such as domestic agriculturalindustrial uses and several other usages (e region envi-ronment of the river has a semiarid climate with the Tigrisbasin having a climatic range from semihumid in theheadwaters to the north to semiarid in the area close towhere it met with the Euphrates river in the southern part ofthe region (e average precipitation in the basin per year isbetween 400 and 600mm even though the annual precip-itation values in the upper and lower parts have been reg-istered as 800 and 150mm respectively Due to the highprecipitation rates in the Zagros Mountains in the east of theTigris basin it has a significantly higher mean precipitationof approximately 300mmyear compared to the Euphratesbasin [37] Precipitation mainly occurs in the basins aroundNovember and April while January to March witnesssnowfall in the mountains (e climatic conditions in thelowlands of Iraq and Syria (semiarid to arid) usually facilitatea considerable loss of water to evapotranspiration in theMesopotamian region (e Tigris basin has an air temper-ature of about minus 35degC during the winter in the ArmenianHighlands and about 40degC on the Jezira plateau during the

summer [38] Baghdad the capital city of Iraq has an av-erage rainfall of about 216mm with rainfall usually expe-rienced from December to February At the capital theTigris river mean river flow is 235m3middotsminus 1 (e maximumweather temperature experienced around is 45degC during thesummer and about 10degC during the winters [39 40]

For the modeling establishment the monthly river flowscale is used to build the predictive model measured at theBaghdad station (see Figure 5) (e historical river flowdatabase is gathered from the USGSData Series 540 [41](estatistical description of the utilized dataset is tabulated inTable 1(e area of the drainage site is 134 000 km2 which is

I1

I2

wn1

wn2

Input weights arerandomly initialized

Output weights areanalytically determined

wnkIn

Inpu

t lay

er

w1k

w12

w11b1

b2

h1β11

f1

fK

o1β1j

β21

β2j

βK1

oj

Out

put

laye

r

bk

βKj

h2

hK

Hidden layer

Figure 3 (e architecture of the single hidden layer ELM model

Input L = (xi ti) | xi isin Rn ti isin Rm i = 1 2 NTraining dataset L consists of a set of N training samplesand their associated output valuesg(x) activation function

H =

(1) procedure ELM(L N g())(2)

(3)

(4)

(5)(6) end procedure

Initialize weights wi and biases bi randomly [ndash1 1] i =

▷ wi is the vector of weights that connect the hidden node

▷ bi is the bias value of the hidden node iCalculate the hidden layer output matrix H

return β

Find output weight β using MP generalized inverse

β = HdaggerT

β = (HT H)ndash1HTT

g(w1x1 + b1) g(wNsimx1 + bNsim)

g(w1xN + b1) g(wNsimxN + bNsim)

1 2 Nsim

Nsim Number of hidden nodesOutput Output weight β

i to all input nodes

▷ The output weights

ELM algorithm for training SLFN

Figure 4 (e extreme learning machine model learningprocedure

4 Complexity

coordinated between the Latitude of 33deg24prime34PrimeN and aLongitude of 44deg20prime32PrimeE Over the past two decades anoticeable negative river flow deterioration was observed forthe Tigris river (is deterioration is associated with severalcauses such as diplomatic issues or climate change Hencethe necessity of providing an accurate and reliable data-intelligence model for this river system is highlyrecommended

4 Proposed SSA-ELM Forecasting Model

(ere are two basic points that should be determined beforeapplying any evolutionary or swarm-based algorithm tooptimize any problem which are the representations of thesolution and the choice of the cost function (ese twopoints are addressed for the proposed SSA-ELM model asfollows

(i) In order to handle an optimization problem using ametaheuristic algorithm the solution of the problemshould be represented properly in the algorithm(at is the way that the solution is encoded as anindividual In the proposed SSA-ELM model eachindividual represents a potential solution as a real

vector I where I consists of two parts inputs weightsand hidden biases as given in equation (10) Allelements of I are generated in [minus 1 1]

I W11 W12 Wnkb1 bk1113858 1113859 (7)

(ii) In all metaheuristic algorithms a cost function isutilized to assess the quality of the generated solu-tions over the course of iterations In this work theroot mean squared error (RMSE) is selected as thecost function which is the most commonly used costfunction in evolutionary extreme learning algo-rithms RMSE can be measured as given in equation(8) (e goal of the algorithm is to find the ELMmodel that has the minimum value of RMSE

RMSE

1113936Nj1 1113936

Kj1 βjg wjxi + bj1113872 1113873 minus ti

m times N

1113971

(8)

(e procedure of the proposed SSA-ELM algorithms canbe described as follows

First the algorithm starts by generating randomly apredefined number N of particles Each of the particlesrepresents a candidate ELM network as explained previously

Iran

N

Turkey

Syria

Iraq

KSA

Case study

Figure 5 (e case study location Baghdad metrological station Tigris river Iraq

Table 1 Descriptive statistics for the mean monthly river flow for the Tigris river at Baghdad (Iraq) (1991ndash2010) [33]

Partition Time period No recordsQ (m3middotsminus 1)

Mean St dev Median Minimum MaximumTraining Jun 1991ndashDec 2007 187 780099 379712 674700 298100 2651000Testing Jan 2007ndashDec 2010 48 489879 136746 445900 331400 936400Complete Jun 1991ndashDec 2010 235 720820 363469 636900 298100 2651000

Complexity 5

(en the fitness of each particle is calculated using the fol-lowing steps

(i) Assign the values of the input weights and hiddenbiases that a given particle carries to a networkstructure

(ii) Calculate the output weights of the ELM networkusing the PM method based on the training part ofthe dataset

(iii) Calculate the RMSE as given in equation (8) basedon the validation part of the dataset

After performing the fitness evaluation step the particlesare sorted according to their fitness values (e best Lparticles that generate ELM networks with lowest RMSEvalues are nominated as leaders (e rest of the particles inthe swarm are considered as followers(e description of theproposed predictive algorithm is displayed in Figure 6

(e positions of the leaders and followers are updatedaccording to equations ((1) and (3)) respectively (eevaluation and update steps are repeated until a predefinednumber of iterations is reached

(e training procedure of the proposed model isestablished using correlated lags computed via the corre-lations statistical approach (ie autocorrelation and partialautocorrelation functions) to detect appropriate input at-tributes to the targeted attribute (see Figure 7) (is processis accomplished in harmony with major researches of theliterature [42 43] Based on the visualization of the corre-lation statistic five lags time of river flow historical infor-mation (attributes variables) influence the one-step aheadriver flow modeling Hence a combination of five scenariosmodel (M1 M2 M5) architecture is constructed asfollows

Model 1 M1 Q(t) fQ(tminus 1)

Model 2 M1 Q(t) fQ(tminus 1) Q(tminus 2)

Model 3 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3)

Model 4 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4)

Model 5 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4) Q(tminus 5)

(9)

where Q(t) is the target river flow step ldquoone month aheadrdquoand Q(tminus 1) Q(tminus 5) are the lag times up to five monthspreviously Here five input combinations are constructedfor the prediction matrix of the SSA-ELM and ELM pre-dictive models

5 Application Results and Analysis

Over the past couple decades massive attention has beenfocused on hydrological time series modeling particularlyfor river flow process (is is owing to the need of an ac-curate and reliable intelligent model that can be imple-mented in real practice Several machine learningapproaches have been proposed to forecast and simulate theriver flow time series As a matter of fact the assumption of

the hydrological time series is originated by stochasticitynonlinearity and redundancy Also the underlying mech-anisms of river flow generation are mostly different fromlow- medium- and high-flow periods In other words theextreme events of flows rely upon the outburst of heavystorm of rainfall events Hence developing a new data-in-telligent model characterized by sophistication to forecast allkind of flow patterns satisfactory is still under processmission for the hydrology engineers In this research theapplication of forecasting the monthly river flow (semiaridenvironment) using the hybrid SSA-ELM model is pre-sented Indeed using a univariate modeling is more chal-lenging for the hydrologist as it is fulfilled for the benefit ofthe catchment that lacks the metrological information

(e perdition skills for the conducted modeling werecarried out using two types of performance indicators in-cluding the goodness (eg (r) and (WI)) and the absoluteerror measures (eg (RMSE) and (MAE)) (e mathe-matical description of these performance indicators is

r 1113936

iNi1 Sobsi minus Sobs1113872 1113873 middot S

predi minus Spred1113872 11138731113960 1113961

1113936iNi1 Sobsi minus Sobs1113872 1113873

2middot

1113936iNi1 S

predi minus Spred1113872 1113873

11139691113970

where minus 1le r minus 1

WI 1 minus1113936

iNi1 Sobsi minus S

predi1113872 1113873

2

1113936iNi1 S

predi minus Sobs

11138681113868111386811138681113868

11138681113868111386811138681113868 + Sobsi minus Sobs

11138681113868111386811138681113868

111386811138681113868111386811138681113874 11138752

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

RMSE

1N

1113944

iN

i1Sobsi minus S

predi1113872 1113873

2

11139741113972

MAE 1N

1113944

iN

i1Sobsi minus S

predi

11138681113868111386811138681113868

11138681113868111386811138681113868

(10)

where Sobsi and Spredi are the observed and forecasted values of

river flow Sobs and Spred are the mean values of river flowand N is the number of the observations of the testing phase

For SSA the size of the population is set to 50 while thenumber of iterations is set to 1000 (ese values were rec-ommended and showed superior results when applied forsolving complex optimization problems in [32] For ELMthe common sigmoid function activation function is used[44] However the number of hidden nodes was system-atically determined by experimenting different numbersstarting from 2 hidden nodes up to 10 with a step of 2 nodeseach time

Table 2 tabulates the efficiency results of the hybrid SSA-ELM model achieved based on the testing data for each inputcombination attribute (e results are validated against thepublished results of ELM support vector regression (SVR) andgeneralized regression neural network (GRNN) which wereconducted in [33] on the same case study In addition Table 2also lists the results of random forests (RF) implemented inWeka the popular suite of machine learning software [45]

6 Complexity

Indeed the motivation of the current study is to improvethe performance of the nontuned predictive model ELMthrough complementing with an evolutionary nature-in-spired novel optimization algorithm (ie SSA) Based on the

investigated results an excellent performance in terms oflower RMSE and MAE values and higher r and WI valuesrespectively when forecasting the present river flow datawas observed (e results exhibited a noticeable

Initialize swarm

Calculate the fitness of each particle

ELM network

Assign a vector of inputs weights and

biases to ELM

Update positions of leaders and followers

No

Return best ELM network with

minimum RMSE

Training data

Yes

Evaluate ELM using RMSE

W1 Wnk b0 bk

Maximum number of iterations

reached

Validation data

Figure 6 Description of the proposed SSA-ELM algorithm

Auto

corr

elat

ion

1

05

0

ndash0520181614121086420

Lag (monthly scale)

(a)

1

05

0

ndash0520181614121086420

Lag (monthly scale)

Part

ial a

utoc

orre

latio

ns

(b)

Figure 7 (e statistical correlation process function of the investigated monthly scale river flow

Complexity 7

enhancement in the prediction skills using the proposedSSA-ELM model in comparison with the benchmark ELM-based model (e improved performance with SSA-ELMattributed to the robustness of the novel evolutionary op-timization method SSA linked to the ELM model (isrobustness may have contributed to the internal functionparameter optimization for each input combination

(e Taylor diagram is another modern graphical pre-sentation to diagnose the predictability of the modelingthrough visualizing the errors from the benchmark observedrecord (is diagram graphically summarizes the modelrsquosefficiency in approximating the accuracy to the observeddata (e Taylor diagram of the hybrid SSA-ELM andclassical ELM is illustrated in Figure 8 (e position of eachmodel appeared on the diagram measures the accuracy ofthat model in accordance to the observed river flow Basedon the diagram presentation the SSA-ELM model has ahigher value of correlation compared to the ELMmodel(eSSA-ELM model approximated well with observed andlocated nearer to the actual record (e RMSE between thepredicted and observed river flow was denoted as propor-tional to the distance to the point of the observation (estandard deviation of the predictive model is proportional tothe radial distance from the origin

For detailed inspection for the performance of the SSA-ELM model over the benchmark ELM model Figure 9presents the relative error (RE) distribution indicator inthe form of boxplot Based on the generated graph thehybrid SSA-ELM performed an acceptable improvementwith average RE between minus 10 and + 10 and over 80 of thetesting period(emaximumREmagnitude recorded on the

overestimation is +35 whereas the underestimation aspectis recorded as minus 25(is is much different from the classicalELM that exceeded the minus 40 in the underestimation aspect

Scatter plots for the ELM and SSA-ELM models arepresented in Figures 10 and 11 respectively (e graphicalpresentation followed the determination coefficient values asit is the square value (e proposed hybrid model wasperformed closer to the ideal line between the observed andforecasted values

It is an excellent fashion of evaluating the currentresearch results by comparing with modern establishedresearches with the regards of hybrid evolutionary nature-inspired algorithms integrated with AI-predictive modelsMost recently the applications of metaheuristic techniqueshave shown an extensive and noticeable optimizationmethods for AI predictive models (is is owing to themain advantages such as flexibility give an optimal so-lution and free from gradient decent drawbacks [32] Inthe last five years there are various successful attempts onhybridizing the AI models with evolutionary optimizationalgorithms in the field of hydrology For instance acombination between the adaptive neurofuzzy system(ANFIS) model with the firefly algorithm (FFA) wasconducted for modeling the roller length of the hydraulicjump [46] An integration of the classical multilayer per-ceptron with FFA for pan evaporation process is given in[47] River flow forecasting for tropical environment wasestablished using the ANFIS-FFA model [48] On the sameregion of the case study implemented in this research thewater quality index prediction model was developed basedon the hybridization of the support vector machine model

Table 2 Performance evaluation for ELM and SSA-ELM SVR GRNN and RF models

Input combinations Models r WI RMSE (m3middotsminus 1) MAE (m3 middot sminus 1)

M1

SSA-ELM 0812 0861 80592 66203ELM 081 082 98272 87012SVR 0761 0802 106749 90905

GRNN 0658 0689 127818 113698RF 0372 0531 189778 131966

M2

SSA-ELM 0809 0850 80516 62185ELM 0817 0834 95571 83899SVR 0715 0769 111228 9514

GRNN 0468 054 125187 103551RF 0373 0549 175765 142347

M3

SSA-ELM 0784 0812 86868 62185ELM 0815 0814 104307 92544SVR 0741 0776 11802 10185

GRNN 0496 054 125187 103551RF 0037 033 28085 22527

M4

SSA-ELM 0800 0855 8784 69069ELM 0803 0826 98432 85117SVR 0728 0769 119539 106585

GRNN 0346 0507 133522 114063RF 0127 039 124648 196745

M5

SSA-ELM 0801 0856 87659 67886ELM 0799 0853 87906 71544SVR 0715 0768 124155 108367

GRNN 0248 0416 13535 112606RF 0222 00001 66876 657031

8 Complexity

with the firefly algorithm [49] All the examples exhibitedearlier on the hybrid models demonstrated an excellentperformance over the AI-based models It is evidenced thatthe present study was performed in harmony with a newera of the hybrid intelligent models in the field of surfacehydrology

For the best knowledge of the practical implementationthe proposed hybrid intelligence SSA-ELM model has thepossibility of the online learning potential As reported inFigure 12 the scheme demonstrated the capacity of the onlineforecasting system in which there is a possibility to bepracticed as a floodingwarning system(e online forecastingsystem starts by obtaining the metrological information (eg

river flow velocity (V) cross section area (A) and riverdischarge (Q)) using sensors device(e data information wastransferred via cricket wireless and digitized into a commaseparated value (CSV) format in which it is readable by anysoftware environment At this stage the learning processbased on the historical records is conducted using the SSA-ELM predictive model (e outcome of the predictive modelwas subjected to a specific range of river flow threshold valuesIf the prediction value is within the threshold the loop of thelearning process is repeated using the new recalled historicalinformation Otherwise if the prediction is out of thethreshold range a notification of the alarm system is inte-grated (global system for mobile communications) that can

ELM

SSA-ELM

Observed

00

01

02

03

04

05

06

07

08

09

095

099

0

50

100

150

200

250

0 50 100 150 200 250

Stan

dard

dev

iatio

n

Standard deviation

Taylor diagram

Correlation

Figure 8 Taylor diagram graphical presentation for the ELM and SSA-ELM predictive models

RE

40

30

20

10

0

ndash10

ndash20

ndash30

ndash40

ndash50

RE SSA-ELMRE ELM

Relative error distribution

Figure 9 Relative error distribution percentage of SSA-ELM and ELM models

Complexity 9

y = 0722x + 1487R2 = 0659

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 07228x + 13008R2 = 06553

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 0694x + 13341R2 = 06147

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 0818x + 10463R2 = 06403

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 4

(d)

y = 0811x + 11097R2 = 06422

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 5

(e)

Figure 10 Scatter plot variation between the observed and forecasted river flow using the SSA-ELM model and for the five inputcombinations

10 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

c1 2eminus (4lL)2

(2)

in which l represents the current iteration and L is a pre-defined number of maximum iterations Note that c1 is avery important variable in the SSA process which controlsthe balance between the exploration and exploitation pro-cesses of the optimization algorithm (e followers updatetheir position accordingly as

xij 05 x

ij minus x

iminus 1j1113872 1113873 (3)

where xij is the ith element of the jth position and ige 2 (e

procedure of the SSA algorithm is presented in Figure 2

22 Extreme LearningMachines Artificial neural networks(ANN) have been used as a universal approximator to dealwith complex problems in a wide spectrum of differentdomains One of these common problems is classificationand regression problems For ANN to be able to predictthe value of the output variable it needs to be trainedbased on a set of examples using a learning algorithm Oneof the recent and successful variants of ANNs is ELMELM is a learning framework for single hidden layerfeedforward neural networks (SLFN) introduced recentlyin [20] (e main merit of the ELM model is the learningprocess convergence speed that emphasizes the capabilityover the classical gradient decent approach for trainingthe SLFN [34] In addition the nonrequirement for al-gorithm tuning parameters is similar to the back-propagation algorithm (e ELM model requires nohuman interactive for tuning [35] Further ELM does notrequire an iterative process to tune the SLFN parametersInstead it randomly modifies the input weights and thehidden biases and then it governs the output weightssystematically using the potential of the MoorendashPenrose(MP) generalized inverse method An illustration of theSLFN-based ELM is shown in Figure 3

(e output of SLFN is trained using a dataset with Ndistinct examples denoted as (xi ti) i 1 2 N where xi isthe input vector and ti is the output vector (e activationfunction g() and k hidden neurons is

fi 1113944k

j1βjg wjxi + bj1113872 1113873 j 1 2 N (4)

where wj denotes the weight vector that links the inputnodes to the jth hidden node bj signifies the bias value of thejth hidden node and βj are the values of the output weightsthat connect the jth hidden node with the output nodes [36]Formula (4) is written compactly as

O H times β (5)

where H is the output matrix of the hidden layer ELMassigns the weights and the biases randomly without theneed to take into consideration the input data ELM de-termines the output weights by solving equation (5) using

Movement direction

Leader salp

Follower salps

Figure 1 Leaders and followers of salps chain movement

Initialize a swarm of slaps xi(i = 1 2 n)While (Termination condition is not met) do

Calculate the fitness of each salp in the swarmSet F as the position of best salpUpdate c1 by equation (2)for (each salp (xi)) do

if (i == 1) then

else

end if

end whileReturn F

end for

Update the leaderrsquos position using equation (1)

Update the followerrsquos position using equation (3)

Check and return salps if they go beyond upper andlower bounds

Pseudocode of the SSA algorithm

Figure 2 Salp Swarm Algorithm procedure

Complexity 3

MP generalized inverse (e output weights can be deter-mined by ELM using the following formula

1113954β Hdagger

times T (6)

(e process of the ELM predictive model is presented inFigure 4

3 Case Study and Site Description

In this study the Tigris river which is located in the Iraqregion is selected for examining the proposed hybrid pre-dictive model (e total length of the river is around1718 km (e river originates from Turkey and flows towardIraq in the southern part Most of the river basin (about 85percentage) is located in Iraq with an approximate catch-ment area of 253000 km2 Tigris and Euphrates rivers arepresenting the chief freshwater resources in Iraq providingwater for multiple usages such as domestic agriculturalindustrial uses and several other usages (e region envi-ronment of the river has a semiarid climate with the Tigrisbasin having a climatic range from semihumid in theheadwaters to the north to semiarid in the area close towhere it met with the Euphrates river in the southern part ofthe region (e average precipitation in the basin per year isbetween 400 and 600mm even though the annual precip-itation values in the upper and lower parts have been reg-istered as 800 and 150mm respectively Due to the highprecipitation rates in the Zagros Mountains in the east of theTigris basin it has a significantly higher mean precipitationof approximately 300mmyear compared to the Euphratesbasin [37] Precipitation mainly occurs in the basins aroundNovember and April while January to March witnesssnowfall in the mountains (e climatic conditions in thelowlands of Iraq and Syria (semiarid to arid) usually facilitatea considerable loss of water to evapotranspiration in theMesopotamian region (e Tigris basin has an air temper-ature of about minus 35degC during the winter in the ArmenianHighlands and about 40degC on the Jezira plateau during the

summer [38] Baghdad the capital city of Iraq has an av-erage rainfall of about 216mm with rainfall usually expe-rienced from December to February At the capital theTigris river mean river flow is 235m3middotsminus 1 (e maximumweather temperature experienced around is 45degC during thesummer and about 10degC during the winters [39 40]

For the modeling establishment the monthly river flowscale is used to build the predictive model measured at theBaghdad station (see Figure 5) (e historical river flowdatabase is gathered from the USGSData Series 540 [41](estatistical description of the utilized dataset is tabulated inTable 1(e area of the drainage site is 134 000 km2 which is

I1

I2

wn1

wn2

Input weights arerandomly initialized

Output weights areanalytically determined

wnkIn

Inpu

t lay

er

w1k

w12

w11b1

b2

h1β11

f1

fK

o1β1j

β21

β2j

βK1

oj

Out

put

laye

r

bk

βKj

h2

hK

Hidden layer

Figure 3 (e architecture of the single hidden layer ELM model

Input L = (xi ti) | xi isin Rn ti isin Rm i = 1 2 NTraining dataset L consists of a set of N training samplesand their associated output valuesg(x) activation function

H =

(1) procedure ELM(L N g())(2)

(3)

(4)

(5)(6) end procedure

Initialize weights wi and biases bi randomly [ndash1 1] i =

▷ wi is the vector of weights that connect the hidden node

▷ bi is the bias value of the hidden node iCalculate the hidden layer output matrix H

return β

Find output weight β using MP generalized inverse

β = HdaggerT

β = (HT H)ndash1HTT

g(w1x1 + b1) g(wNsimx1 + bNsim)

g(w1xN + b1) g(wNsimxN + bNsim)

1 2 Nsim

Nsim Number of hidden nodesOutput Output weight β

i to all input nodes

▷ The output weights

ELM algorithm for training SLFN

Figure 4 (e extreme learning machine model learningprocedure

4 Complexity

coordinated between the Latitude of 33deg24prime34PrimeN and aLongitude of 44deg20prime32PrimeE Over the past two decades anoticeable negative river flow deterioration was observed forthe Tigris river (is deterioration is associated with severalcauses such as diplomatic issues or climate change Hencethe necessity of providing an accurate and reliable data-intelligence model for this river system is highlyrecommended

4 Proposed SSA-ELM Forecasting Model

(ere are two basic points that should be determined beforeapplying any evolutionary or swarm-based algorithm tooptimize any problem which are the representations of thesolution and the choice of the cost function (ese twopoints are addressed for the proposed SSA-ELM model asfollows

(i) In order to handle an optimization problem using ametaheuristic algorithm the solution of the problemshould be represented properly in the algorithm(at is the way that the solution is encoded as anindividual In the proposed SSA-ELM model eachindividual represents a potential solution as a real

vector I where I consists of two parts inputs weightsand hidden biases as given in equation (10) Allelements of I are generated in [minus 1 1]

I W11 W12 Wnkb1 bk1113858 1113859 (7)

(ii) In all metaheuristic algorithms a cost function isutilized to assess the quality of the generated solu-tions over the course of iterations In this work theroot mean squared error (RMSE) is selected as thecost function which is the most commonly used costfunction in evolutionary extreme learning algo-rithms RMSE can be measured as given in equation(8) (e goal of the algorithm is to find the ELMmodel that has the minimum value of RMSE

RMSE

1113936Nj1 1113936

Kj1 βjg wjxi + bj1113872 1113873 minus ti

m times N

1113971

(8)

(e procedure of the proposed SSA-ELM algorithms canbe described as follows

First the algorithm starts by generating randomly apredefined number N of particles Each of the particlesrepresents a candidate ELM network as explained previously

Iran

N

Turkey

Syria

Iraq

KSA

Case study

Figure 5 (e case study location Baghdad metrological station Tigris river Iraq

Table 1 Descriptive statistics for the mean monthly river flow for the Tigris river at Baghdad (Iraq) (1991ndash2010) [33]

Partition Time period No recordsQ (m3middotsminus 1)

Mean St dev Median Minimum MaximumTraining Jun 1991ndashDec 2007 187 780099 379712 674700 298100 2651000Testing Jan 2007ndashDec 2010 48 489879 136746 445900 331400 936400Complete Jun 1991ndashDec 2010 235 720820 363469 636900 298100 2651000

Complexity 5

(en the fitness of each particle is calculated using the fol-lowing steps

(i) Assign the values of the input weights and hiddenbiases that a given particle carries to a networkstructure

(ii) Calculate the output weights of the ELM networkusing the PM method based on the training part ofthe dataset

(iii) Calculate the RMSE as given in equation (8) basedon the validation part of the dataset

After performing the fitness evaluation step the particlesare sorted according to their fitness values (e best Lparticles that generate ELM networks with lowest RMSEvalues are nominated as leaders (e rest of the particles inthe swarm are considered as followers(e description of theproposed predictive algorithm is displayed in Figure 6

(e positions of the leaders and followers are updatedaccording to equations ((1) and (3)) respectively (eevaluation and update steps are repeated until a predefinednumber of iterations is reached

(e training procedure of the proposed model isestablished using correlated lags computed via the corre-lations statistical approach (ie autocorrelation and partialautocorrelation functions) to detect appropriate input at-tributes to the targeted attribute (see Figure 7) (is processis accomplished in harmony with major researches of theliterature [42 43] Based on the visualization of the corre-lation statistic five lags time of river flow historical infor-mation (attributes variables) influence the one-step aheadriver flow modeling Hence a combination of five scenariosmodel (M1 M2 M5) architecture is constructed asfollows

Model 1 M1 Q(t) fQ(tminus 1)

Model 2 M1 Q(t) fQ(tminus 1) Q(tminus 2)

Model 3 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3)

Model 4 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4)

Model 5 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4) Q(tminus 5)

(9)

where Q(t) is the target river flow step ldquoone month aheadrdquoand Q(tminus 1) Q(tminus 5) are the lag times up to five monthspreviously Here five input combinations are constructedfor the prediction matrix of the SSA-ELM and ELM pre-dictive models

5 Application Results and Analysis

Over the past couple decades massive attention has beenfocused on hydrological time series modeling particularlyfor river flow process (is is owing to the need of an ac-curate and reliable intelligent model that can be imple-mented in real practice Several machine learningapproaches have been proposed to forecast and simulate theriver flow time series As a matter of fact the assumption of

the hydrological time series is originated by stochasticitynonlinearity and redundancy Also the underlying mech-anisms of river flow generation are mostly different fromlow- medium- and high-flow periods In other words theextreme events of flows rely upon the outburst of heavystorm of rainfall events Hence developing a new data-in-telligent model characterized by sophistication to forecast allkind of flow patterns satisfactory is still under processmission for the hydrology engineers In this research theapplication of forecasting the monthly river flow (semiaridenvironment) using the hybrid SSA-ELM model is pre-sented Indeed using a univariate modeling is more chal-lenging for the hydrologist as it is fulfilled for the benefit ofthe catchment that lacks the metrological information

(e perdition skills for the conducted modeling werecarried out using two types of performance indicators in-cluding the goodness (eg (r) and (WI)) and the absoluteerror measures (eg (RMSE) and (MAE)) (e mathe-matical description of these performance indicators is

r 1113936

iNi1 Sobsi minus Sobs1113872 1113873 middot S

predi minus Spred1113872 11138731113960 1113961

1113936iNi1 Sobsi minus Sobs1113872 1113873

2middot

1113936iNi1 S

predi minus Spred1113872 1113873

11139691113970

where minus 1le r minus 1

WI 1 minus1113936

iNi1 Sobsi minus S

predi1113872 1113873

2

1113936iNi1 S

predi minus Sobs

11138681113868111386811138681113868

11138681113868111386811138681113868 + Sobsi minus Sobs

11138681113868111386811138681113868

111386811138681113868111386811138681113874 11138752

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

RMSE

1N

1113944

iN

i1Sobsi minus S

predi1113872 1113873

2

11139741113972

MAE 1N

1113944

iN

i1Sobsi minus S

predi

11138681113868111386811138681113868

11138681113868111386811138681113868

(10)

where Sobsi and Spredi are the observed and forecasted values of

river flow Sobs and Spred are the mean values of river flowand N is the number of the observations of the testing phase

For SSA the size of the population is set to 50 while thenumber of iterations is set to 1000 (ese values were rec-ommended and showed superior results when applied forsolving complex optimization problems in [32] For ELMthe common sigmoid function activation function is used[44] However the number of hidden nodes was system-atically determined by experimenting different numbersstarting from 2 hidden nodes up to 10 with a step of 2 nodeseach time

Table 2 tabulates the efficiency results of the hybrid SSA-ELM model achieved based on the testing data for each inputcombination attribute (e results are validated against thepublished results of ELM support vector regression (SVR) andgeneralized regression neural network (GRNN) which wereconducted in [33] on the same case study In addition Table 2also lists the results of random forests (RF) implemented inWeka the popular suite of machine learning software [45]

6 Complexity

Indeed the motivation of the current study is to improvethe performance of the nontuned predictive model ELMthrough complementing with an evolutionary nature-in-spired novel optimization algorithm (ie SSA) Based on the

investigated results an excellent performance in terms oflower RMSE and MAE values and higher r and WI valuesrespectively when forecasting the present river flow datawas observed (e results exhibited a noticeable

Initialize swarm

Calculate the fitness of each particle

ELM network

Assign a vector of inputs weights and

biases to ELM

Update positions of leaders and followers

No

Return best ELM network with

minimum RMSE

Training data

Yes

Evaluate ELM using RMSE

W1 Wnk b0 bk

Maximum number of iterations

reached

Validation data

Figure 6 Description of the proposed SSA-ELM algorithm

Auto

corr

elat

ion

1

05

0

ndash0520181614121086420

Lag (monthly scale)

(a)

1

05

0

ndash0520181614121086420

Lag (monthly scale)

Part

ial a

utoc

orre

latio

ns

(b)

Figure 7 (e statistical correlation process function of the investigated monthly scale river flow

Complexity 7

enhancement in the prediction skills using the proposedSSA-ELM model in comparison with the benchmark ELM-based model (e improved performance with SSA-ELMattributed to the robustness of the novel evolutionary op-timization method SSA linked to the ELM model (isrobustness may have contributed to the internal functionparameter optimization for each input combination

(e Taylor diagram is another modern graphical pre-sentation to diagnose the predictability of the modelingthrough visualizing the errors from the benchmark observedrecord (is diagram graphically summarizes the modelrsquosefficiency in approximating the accuracy to the observeddata (e Taylor diagram of the hybrid SSA-ELM andclassical ELM is illustrated in Figure 8 (e position of eachmodel appeared on the diagram measures the accuracy ofthat model in accordance to the observed river flow Basedon the diagram presentation the SSA-ELM model has ahigher value of correlation compared to the ELMmodel(eSSA-ELM model approximated well with observed andlocated nearer to the actual record (e RMSE between thepredicted and observed river flow was denoted as propor-tional to the distance to the point of the observation (estandard deviation of the predictive model is proportional tothe radial distance from the origin

For detailed inspection for the performance of the SSA-ELM model over the benchmark ELM model Figure 9presents the relative error (RE) distribution indicator inthe form of boxplot Based on the generated graph thehybrid SSA-ELM performed an acceptable improvementwith average RE between minus 10 and + 10 and over 80 of thetesting period(emaximumREmagnitude recorded on the

overestimation is +35 whereas the underestimation aspectis recorded as minus 25(is is much different from the classicalELM that exceeded the minus 40 in the underestimation aspect

Scatter plots for the ELM and SSA-ELM models arepresented in Figures 10 and 11 respectively (e graphicalpresentation followed the determination coefficient values asit is the square value (e proposed hybrid model wasperformed closer to the ideal line between the observed andforecasted values

It is an excellent fashion of evaluating the currentresearch results by comparing with modern establishedresearches with the regards of hybrid evolutionary nature-inspired algorithms integrated with AI-predictive modelsMost recently the applications of metaheuristic techniqueshave shown an extensive and noticeable optimizationmethods for AI predictive models (is is owing to themain advantages such as flexibility give an optimal so-lution and free from gradient decent drawbacks [32] Inthe last five years there are various successful attempts onhybridizing the AI models with evolutionary optimizationalgorithms in the field of hydrology For instance acombination between the adaptive neurofuzzy system(ANFIS) model with the firefly algorithm (FFA) wasconducted for modeling the roller length of the hydraulicjump [46] An integration of the classical multilayer per-ceptron with FFA for pan evaporation process is given in[47] River flow forecasting for tropical environment wasestablished using the ANFIS-FFA model [48] On the sameregion of the case study implemented in this research thewater quality index prediction model was developed basedon the hybridization of the support vector machine model

Table 2 Performance evaluation for ELM and SSA-ELM SVR GRNN and RF models

Input combinations Models r WI RMSE (m3middotsminus 1) MAE (m3 middot sminus 1)

M1

SSA-ELM 0812 0861 80592 66203ELM 081 082 98272 87012SVR 0761 0802 106749 90905

GRNN 0658 0689 127818 113698RF 0372 0531 189778 131966

M2

SSA-ELM 0809 0850 80516 62185ELM 0817 0834 95571 83899SVR 0715 0769 111228 9514

GRNN 0468 054 125187 103551RF 0373 0549 175765 142347

M3

SSA-ELM 0784 0812 86868 62185ELM 0815 0814 104307 92544SVR 0741 0776 11802 10185

GRNN 0496 054 125187 103551RF 0037 033 28085 22527

M4

SSA-ELM 0800 0855 8784 69069ELM 0803 0826 98432 85117SVR 0728 0769 119539 106585

GRNN 0346 0507 133522 114063RF 0127 039 124648 196745

M5

SSA-ELM 0801 0856 87659 67886ELM 0799 0853 87906 71544SVR 0715 0768 124155 108367

GRNN 0248 0416 13535 112606RF 0222 00001 66876 657031

8 Complexity

with the firefly algorithm [49] All the examples exhibitedearlier on the hybrid models demonstrated an excellentperformance over the AI-based models It is evidenced thatthe present study was performed in harmony with a newera of the hybrid intelligent models in the field of surfacehydrology

For the best knowledge of the practical implementationthe proposed hybrid intelligence SSA-ELM model has thepossibility of the online learning potential As reported inFigure 12 the scheme demonstrated the capacity of the onlineforecasting system in which there is a possibility to bepracticed as a floodingwarning system(e online forecastingsystem starts by obtaining the metrological information (eg

river flow velocity (V) cross section area (A) and riverdischarge (Q)) using sensors device(e data information wastransferred via cricket wireless and digitized into a commaseparated value (CSV) format in which it is readable by anysoftware environment At this stage the learning processbased on the historical records is conducted using the SSA-ELM predictive model (e outcome of the predictive modelwas subjected to a specific range of river flow threshold valuesIf the prediction value is within the threshold the loop of thelearning process is repeated using the new recalled historicalinformation Otherwise if the prediction is out of thethreshold range a notification of the alarm system is inte-grated (global system for mobile communications) that can

ELM

SSA-ELM

Observed

00

01

02

03

04

05

06

07

08

09

095

099

0

50

100

150

200

250

0 50 100 150 200 250

Stan

dard

dev

iatio

n

Standard deviation

Taylor diagram

Correlation

Figure 8 Taylor diagram graphical presentation for the ELM and SSA-ELM predictive models

RE

40

30

20

10

0

ndash10

ndash20

ndash30

ndash40

ndash50

RE SSA-ELMRE ELM

Relative error distribution

Figure 9 Relative error distribution percentage of SSA-ELM and ELM models

Complexity 9

y = 0722x + 1487R2 = 0659

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 07228x + 13008R2 = 06553

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 0694x + 13341R2 = 06147

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 0818x + 10463R2 = 06403

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 4

(d)

y = 0811x + 11097R2 = 06422

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 5

(e)

Figure 10 Scatter plot variation between the observed and forecasted river flow using the SSA-ELM model and for the five inputcombinations

10 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

MP generalized inverse (e output weights can be deter-mined by ELM using the following formula

1113954β Hdagger

times T (6)

(e process of the ELM predictive model is presented inFigure 4

3 Case Study and Site Description

In this study the Tigris river which is located in the Iraqregion is selected for examining the proposed hybrid pre-dictive model (e total length of the river is around1718 km (e river originates from Turkey and flows towardIraq in the southern part Most of the river basin (about 85percentage) is located in Iraq with an approximate catch-ment area of 253000 km2 Tigris and Euphrates rivers arepresenting the chief freshwater resources in Iraq providingwater for multiple usages such as domestic agriculturalindustrial uses and several other usages (e region envi-ronment of the river has a semiarid climate with the Tigrisbasin having a climatic range from semihumid in theheadwaters to the north to semiarid in the area close towhere it met with the Euphrates river in the southern part ofthe region (e average precipitation in the basin per year isbetween 400 and 600mm even though the annual precip-itation values in the upper and lower parts have been reg-istered as 800 and 150mm respectively Due to the highprecipitation rates in the Zagros Mountains in the east of theTigris basin it has a significantly higher mean precipitationof approximately 300mmyear compared to the Euphratesbasin [37] Precipitation mainly occurs in the basins aroundNovember and April while January to March witnesssnowfall in the mountains (e climatic conditions in thelowlands of Iraq and Syria (semiarid to arid) usually facilitatea considerable loss of water to evapotranspiration in theMesopotamian region (e Tigris basin has an air temper-ature of about minus 35degC during the winter in the ArmenianHighlands and about 40degC on the Jezira plateau during the

summer [38] Baghdad the capital city of Iraq has an av-erage rainfall of about 216mm with rainfall usually expe-rienced from December to February At the capital theTigris river mean river flow is 235m3middotsminus 1 (e maximumweather temperature experienced around is 45degC during thesummer and about 10degC during the winters [39 40]

For the modeling establishment the monthly river flowscale is used to build the predictive model measured at theBaghdad station (see Figure 5) (e historical river flowdatabase is gathered from the USGSData Series 540 [41](estatistical description of the utilized dataset is tabulated inTable 1(e area of the drainage site is 134 000 km2 which is

I1

I2

wn1

wn2

Input weights arerandomly initialized

Output weights areanalytically determined

wnkIn

Inpu

t lay

er

w1k

w12

w11b1

b2

h1β11

f1

fK

o1β1j

β21

β2j

βK1

oj

Out

put

laye

r

bk

βKj

h2

hK

Hidden layer

Figure 3 (e architecture of the single hidden layer ELM model

Input L = (xi ti) | xi isin Rn ti isin Rm i = 1 2 NTraining dataset L consists of a set of N training samplesand their associated output valuesg(x) activation function

H =

(1) procedure ELM(L N g())(2)

(3)

(4)

(5)(6) end procedure

Initialize weights wi and biases bi randomly [ndash1 1] i =

▷ wi is the vector of weights that connect the hidden node

▷ bi is the bias value of the hidden node iCalculate the hidden layer output matrix H

return β

Find output weight β using MP generalized inverse

β = HdaggerT

β = (HT H)ndash1HTT

g(w1x1 + b1) g(wNsimx1 + bNsim)

g(w1xN + b1) g(wNsimxN + bNsim)

1 2 Nsim

Nsim Number of hidden nodesOutput Output weight β

i to all input nodes

▷ The output weights

ELM algorithm for training SLFN

Figure 4 (e extreme learning machine model learningprocedure

4 Complexity

coordinated between the Latitude of 33deg24prime34PrimeN and aLongitude of 44deg20prime32PrimeE Over the past two decades anoticeable negative river flow deterioration was observed forthe Tigris river (is deterioration is associated with severalcauses such as diplomatic issues or climate change Hencethe necessity of providing an accurate and reliable data-intelligence model for this river system is highlyrecommended

4 Proposed SSA-ELM Forecasting Model

(ere are two basic points that should be determined beforeapplying any evolutionary or swarm-based algorithm tooptimize any problem which are the representations of thesolution and the choice of the cost function (ese twopoints are addressed for the proposed SSA-ELM model asfollows

(i) In order to handle an optimization problem using ametaheuristic algorithm the solution of the problemshould be represented properly in the algorithm(at is the way that the solution is encoded as anindividual In the proposed SSA-ELM model eachindividual represents a potential solution as a real

vector I where I consists of two parts inputs weightsand hidden biases as given in equation (10) Allelements of I are generated in [minus 1 1]

I W11 W12 Wnkb1 bk1113858 1113859 (7)

(ii) In all metaheuristic algorithms a cost function isutilized to assess the quality of the generated solu-tions over the course of iterations In this work theroot mean squared error (RMSE) is selected as thecost function which is the most commonly used costfunction in evolutionary extreme learning algo-rithms RMSE can be measured as given in equation(8) (e goal of the algorithm is to find the ELMmodel that has the minimum value of RMSE

RMSE

1113936Nj1 1113936

Kj1 βjg wjxi + bj1113872 1113873 minus ti

m times N

1113971

(8)

(e procedure of the proposed SSA-ELM algorithms canbe described as follows

First the algorithm starts by generating randomly apredefined number N of particles Each of the particlesrepresents a candidate ELM network as explained previously

Iran

N

Turkey

Syria

Iraq

KSA

Case study

Figure 5 (e case study location Baghdad metrological station Tigris river Iraq

Table 1 Descriptive statistics for the mean monthly river flow for the Tigris river at Baghdad (Iraq) (1991ndash2010) [33]

Partition Time period No recordsQ (m3middotsminus 1)

Mean St dev Median Minimum MaximumTraining Jun 1991ndashDec 2007 187 780099 379712 674700 298100 2651000Testing Jan 2007ndashDec 2010 48 489879 136746 445900 331400 936400Complete Jun 1991ndashDec 2010 235 720820 363469 636900 298100 2651000

Complexity 5

(en the fitness of each particle is calculated using the fol-lowing steps

(i) Assign the values of the input weights and hiddenbiases that a given particle carries to a networkstructure

(ii) Calculate the output weights of the ELM networkusing the PM method based on the training part ofthe dataset

(iii) Calculate the RMSE as given in equation (8) basedon the validation part of the dataset

After performing the fitness evaluation step the particlesare sorted according to their fitness values (e best Lparticles that generate ELM networks with lowest RMSEvalues are nominated as leaders (e rest of the particles inthe swarm are considered as followers(e description of theproposed predictive algorithm is displayed in Figure 6

(e positions of the leaders and followers are updatedaccording to equations ((1) and (3)) respectively (eevaluation and update steps are repeated until a predefinednumber of iterations is reached

(e training procedure of the proposed model isestablished using correlated lags computed via the corre-lations statistical approach (ie autocorrelation and partialautocorrelation functions) to detect appropriate input at-tributes to the targeted attribute (see Figure 7) (is processis accomplished in harmony with major researches of theliterature [42 43] Based on the visualization of the corre-lation statistic five lags time of river flow historical infor-mation (attributes variables) influence the one-step aheadriver flow modeling Hence a combination of five scenariosmodel (M1 M2 M5) architecture is constructed asfollows

Model 1 M1 Q(t) fQ(tminus 1)

Model 2 M1 Q(t) fQ(tminus 1) Q(tminus 2)

Model 3 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3)

Model 4 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4)

Model 5 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4) Q(tminus 5)

(9)

where Q(t) is the target river flow step ldquoone month aheadrdquoand Q(tminus 1) Q(tminus 5) are the lag times up to five monthspreviously Here five input combinations are constructedfor the prediction matrix of the SSA-ELM and ELM pre-dictive models

5 Application Results and Analysis

Over the past couple decades massive attention has beenfocused on hydrological time series modeling particularlyfor river flow process (is is owing to the need of an ac-curate and reliable intelligent model that can be imple-mented in real practice Several machine learningapproaches have been proposed to forecast and simulate theriver flow time series As a matter of fact the assumption of

the hydrological time series is originated by stochasticitynonlinearity and redundancy Also the underlying mech-anisms of river flow generation are mostly different fromlow- medium- and high-flow periods In other words theextreme events of flows rely upon the outburst of heavystorm of rainfall events Hence developing a new data-in-telligent model characterized by sophistication to forecast allkind of flow patterns satisfactory is still under processmission for the hydrology engineers In this research theapplication of forecasting the monthly river flow (semiaridenvironment) using the hybrid SSA-ELM model is pre-sented Indeed using a univariate modeling is more chal-lenging for the hydrologist as it is fulfilled for the benefit ofthe catchment that lacks the metrological information

(e perdition skills for the conducted modeling werecarried out using two types of performance indicators in-cluding the goodness (eg (r) and (WI)) and the absoluteerror measures (eg (RMSE) and (MAE)) (e mathe-matical description of these performance indicators is

r 1113936

iNi1 Sobsi minus Sobs1113872 1113873 middot S

predi minus Spred1113872 11138731113960 1113961

1113936iNi1 Sobsi minus Sobs1113872 1113873

2middot

1113936iNi1 S

predi minus Spred1113872 1113873

11139691113970

where minus 1le r minus 1

WI 1 minus1113936

iNi1 Sobsi minus S

predi1113872 1113873

2

1113936iNi1 S

predi minus Sobs

11138681113868111386811138681113868

11138681113868111386811138681113868 + Sobsi minus Sobs

11138681113868111386811138681113868

111386811138681113868111386811138681113874 11138752

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

RMSE

1N

1113944

iN

i1Sobsi minus S

predi1113872 1113873

2

11139741113972

MAE 1N

1113944

iN

i1Sobsi minus S

predi

11138681113868111386811138681113868

11138681113868111386811138681113868

(10)

where Sobsi and Spredi are the observed and forecasted values of

river flow Sobs and Spred are the mean values of river flowand N is the number of the observations of the testing phase

For SSA the size of the population is set to 50 while thenumber of iterations is set to 1000 (ese values were rec-ommended and showed superior results when applied forsolving complex optimization problems in [32] For ELMthe common sigmoid function activation function is used[44] However the number of hidden nodes was system-atically determined by experimenting different numbersstarting from 2 hidden nodes up to 10 with a step of 2 nodeseach time

Table 2 tabulates the efficiency results of the hybrid SSA-ELM model achieved based on the testing data for each inputcombination attribute (e results are validated against thepublished results of ELM support vector regression (SVR) andgeneralized regression neural network (GRNN) which wereconducted in [33] on the same case study In addition Table 2also lists the results of random forests (RF) implemented inWeka the popular suite of machine learning software [45]

6 Complexity

Indeed the motivation of the current study is to improvethe performance of the nontuned predictive model ELMthrough complementing with an evolutionary nature-in-spired novel optimization algorithm (ie SSA) Based on the

investigated results an excellent performance in terms oflower RMSE and MAE values and higher r and WI valuesrespectively when forecasting the present river flow datawas observed (e results exhibited a noticeable

Initialize swarm

Calculate the fitness of each particle

ELM network

Assign a vector of inputs weights and

biases to ELM

Update positions of leaders and followers

No

Return best ELM network with

minimum RMSE

Training data

Yes

Evaluate ELM using RMSE

W1 Wnk b0 bk

Maximum number of iterations

reached

Validation data

Figure 6 Description of the proposed SSA-ELM algorithm

Auto

corr

elat

ion

1

05

0

ndash0520181614121086420

Lag (monthly scale)

(a)

1

05

0

ndash0520181614121086420

Lag (monthly scale)

Part

ial a

utoc

orre

latio

ns

(b)

Figure 7 (e statistical correlation process function of the investigated monthly scale river flow

Complexity 7

enhancement in the prediction skills using the proposedSSA-ELM model in comparison with the benchmark ELM-based model (e improved performance with SSA-ELMattributed to the robustness of the novel evolutionary op-timization method SSA linked to the ELM model (isrobustness may have contributed to the internal functionparameter optimization for each input combination

(e Taylor diagram is another modern graphical pre-sentation to diagnose the predictability of the modelingthrough visualizing the errors from the benchmark observedrecord (is diagram graphically summarizes the modelrsquosefficiency in approximating the accuracy to the observeddata (e Taylor diagram of the hybrid SSA-ELM andclassical ELM is illustrated in Figure 8 (e position of eachmodel appeared on the diagram measures the accuracy ofthat model in accordance to the observed river flow Basedon the diagram presentation the SSA-ELM model has ahigher value of correlation compared to the ELMmodel(eSSA-ELM model approximated well with observed andlocated nearer to the actual record (e RMSE between thepredicted and observed river flow was denoted as propor-tional to the distance to the point of the observation (estandard deviation of the predictive model is proportional tothe radial distance from the origin

For detailed inspection for the performance of the SSA-ELM model over the benchmark ELM model Figure 9presents the relative error (RE) distribution indicator inthe form of boxplot Based on the generated graph thehybrid SSA-ELM performed an acceptable improvementwith average RE between minus 10 and + 10 and over 80 of thetesting period(emaximumREmagnitude recorded on the

overestimation is +35 whereas the underestimation aspectis recorded as minus 25(is is much different from the classicalELM that exceeded the minus 40 in the underestimation aspect

Scatter plots for the ELM and SSA-ELM models arepresented in Figures 10 and 11 respectively (e graphicalpresentation followed the determination coefficient values asit is the square value (e proposed hybrid model wasperformed closer to the ideal line between the observed andforecasted values

It is an excellent fashion of evaluating the currentresearch results by comparing with modern establishedresearches with the regards of hybrid evolutionary nature-inspired algorithms integrated with AI-predictive modelsMost recently the applications of metaheuristic techniqueshave shown an extensive and noticeable optimizationmethods for AI predictive models (is is owing to themain advantages such as flexibility give an optimal so-lution and free from gradient decent drawbacks [32] Inthe last five years there are various successful attempts onhybridizing the AI models with evolutionary optimizationalgorithms in the field of hydrology For instance acombination between the adaptive neurofuzzy system(ANFIS) model with the firefly algorithm (FFA) wasconducted for modeling the roller length of the hydraulicjump [46] An integration of the classical multilayer per-ceptron with FFA for pan evaporation process is given in[47] River flow forecasting for tropical environment wasestablished using the ANFIS-FFA model [48] On the sameregion of the case study implemented in this research thewater quality index prediction model was developed basedon the hybridization of the support vector machine model

Table 2 Performance evaluation for ELM and SSA-ELM SVR GRNN and RF models

Input combinations Models r WI RMSE (m3middotsminus 1) MAE (m3 middot sminus 1)

M1

SSA-ELM 0812 0861 80592 66203ELM 081 082 98272 87012SVR 0761 0802 106749 90905

GRNN 0658 0689 127818 113698RF 0372 0531 189778 131966

M2

SSA-ELM 0809 0850 80516 62185ELM 0817 0834 95571 83899SVR 0715 0769 111228 9514

GRNN 0468 054 125187 103551RF 0373 0549 175765 142347

M3

SSA-ELM 0784 0812 86868 62185ELM 0815 0814 104307 92544SVR 0741 0776 11802 10185

GRNN 0496 054 125187 103551RF 0037 033 28085 22527

M4

SSA-ELM 0800 0855 8784 69069ELM 0803 0826 98432 85117SVR 0728 0769 119539 106585

GRNN 0346 0507 133522 114063RF 0127 039 124648 196745

M5

SSA-ELM 0801 0856 87659 67886ELM 0799 0853 87906 71544SVR 0715 0768 124155 108367

GRNN 0248 0416 13535 112606RF 0222 00001 66876 657031

8 Complexity

with the firefly algorithm [49] All the examples exhibitedearlier on the hybrid models demonstrated an excellentperformance over the AI-based models It is evidenced thatthe present study was performed in harmony with a newera of the hybrid intelligent models in the field of surfacehydrology

For the best knowledge of the practical implementationthe proposed hybrid intelligence SSA-ELM model has thepossibility of the online learning potential As reported inFigure 12 the scheme demonstrated the capacity of the onlineforecasting system in which there is a possibility to bepracticed as a floodingwarning system(e online forecastingsystem starts by obtaining the metrological information (eg

river flow velocity (V) cross section area (A) and riverdischarge (Q)) using sensors device(e data information wastransferred via cricket wireless and digitized into a commaseparated value (CSV) format in which it is readable by anysoftware environment At this stage the learning processbased on the historical records is conducted using the SSA-ELM predictive model (e outcome of the predictive modelwas subjected to a specific range of river flow threshold valuesIf the prediction value is within the threshold the loop of thelearning process is repeated using the new recalled historicalinformation Otherwise if the prediction is out of thethreshold range a notification of the alarm system is inte-grated (global system for mobile communications) that can

ELM

SSA-ELM

Observed

00

01

02

03

04

05

06

07

08

09

095

099

0

50

100

150

200

250

0 50 100 150 200 250

Stan

dard

dev

iatio

n

Standard deviation

Taylor diagram

Correlation

Figure 8 Taylor diagram graphical presentation for the ELM and SSA-ELM predictive models

RE

40

30

20

10

0

ndash10

ndash20

ndash30

ndash40

ndash50

RE SSA-ELMRE ELM

Relative error distribution

Figure 9 Relative error distribution percentage of SSA-ELM and ELM models

Complexity 9

y = 0722x + 1487R2 = 0659

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 07228x + 13008R2 = 06553

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 0694x + 13341R2 = 06147

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 0818x + 10463R2 = 06403

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 4

(d)

y = 0811x + 11097R2 = 06422

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 5

(e)

Figure 10 Scatter plot variation between the observed and forecasted river flow using the SSA-ELM model and for the five inputcombinations

10 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

coordinated between the Latitude of 33deg24prime34PrimeN and aLongitude of 44deg20prime32PrimeE Over the past two decades anoticeable negative river flow deterioration was observed forthe Tigris river (is deterioration is associated with severalcauses such as diplomatic issues or climate change Hencethe necessity of providing an accurate and reliable data-intelligence model for this river system is highlyrecommended

4 Proposed SSA-ELM Forecasting Model

(ere are two basic points that should be determined beforeapplying any evolutionary or swarm-based algorithm tooptimize any problem which are the representations of thesolution and the choice of the cost function (ese twopoints are addressed for the proposed SSA-ELM model asfollows

(i) In order to handle an optimization problem using ametaheuristic algorithm the solution of the problemshould be represented properly in the algorithm(at is the way that the solution is encoded as anindividual In the proposed SSA-ELM model eachindividual represents a potential solution as a real

vector I where I consists of two parts inputs weightsand hidden biases as given in equation (10) Allelements of I are generated in [minus 1 1]

I W11 W12 Wnkb1 bk1113858 1113859 (7)

(ii) In all metaheuristic algorithms a cost function isutilized to assess the quality of the generated solu-tions over the course of iterations In this work theroot mean squared error (RMSE) is selected as thecost function which is the most commonly used costfunction in evolutionary extreme learning algo-rithms RMSE can be measured as given in equation(8) (e goal of the algorithm is to find the ELMmodel that has the minimum value of RMSE

RMSE

1113936Nj1 1113936

Kj1 βjg wjxi + bj1113872 1113873 minus ti

m times N

1113971

(8)

(e procedure of the proposed SSA-ELM algorithms canbe described as follows

First the algorithm starts by generating randomly apredefined number N of particles Each of the particlesrepresents a candidate ELM network as explained previously

Iran

N

Turkey

Syria

Iraq

KSA

Case study

Figure 5 (e case study location Baghdad metrological station Tigris river Iraq

Table 1 Descriptive statistics for the mean monthly river flow for the Tigris river at Baghdad (Iraq) (1991ndash2010) [33]

Partition Time period No recordsQ (m3middotsminus 1)

Mean St dev Median Minimum MaximumTraining Jun 1991ndashDec 2007 187 780099 379712 674700 298100 2651000Testing Jan 2007ndashDec 2010 48 489879 136746 445900 331400 936400Complete Jun 1991ndashDec 2010 235 720820 363469 636900 298100 2651000

Complexity 5

(en the fitness of each particle is calculated using the fol-lowing steps

(i) Assign the values of the input weights and hiddenbiases that a given particle carries to a networkstructure

(ii) Calculate the output weights of the ELM networkusing the PM method based on the training part ofthe dataset

(iii) Calculate the RMSE as given in equation (8) basedon the validation part of the dataset

After performing the fitness evaluation step the particlesare sorted according to their fitness values (e best Lparticles that generate ELM networks with lowest RMSEvalues are nominated as leaders (e rest of the particles inthe swarm are considered as followers(e description of theproposed predictive algorithm is displayed in Figure 6

(e positions of the leaders and followers are updatedaccording to equations ((1) and (3)) respectively (eevaluation and update steps are repeated until a predefinednumber of iterations is reached

(e training procedure of the proposed model isestablished using correlated lags computed via the corre-lations statistical approach (ie autocorrelation and partialautocorrelation functions) to detect appropriate input at-tributes to the targeted attribute (see Figure 7) (is processis accomplished in harmony with major researches of theliterature [42 43] Based on the visualization of the corre-lation statistic five lags time of river flow historical infor-mation (attributes variables) influence the one-step aheadriver flow modeling Hence a combination of five scenariosmodel (M1 M2 M5) architecture is constructed asfollows

Model 1 M1 Q(t) fQ(tminus 1)

Model 2 M1 Q(t) fQ(tminus 1) Q(tminus 2)

Model 3 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3)

Model 4 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4)

Model 5 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4) Q(tminus 5)

(9)

where Q(t) is the target river flow step ldquoone month aheadrdquoand Q(tminus 1) Q(tminus 5) are the lag times up to five monthspreviously Here five input combinations are constructedfor the prediction matrix of the SSA-ELM and ELM pre-dictive models

5 Application Results and Analysis

Over the past couple decades massive attention has beenfocused on hydrological time series modeling particularlyfor river flow process (is is owing to the need of an ac-curate and reliable intelligent model that can be imple-mented in real practice Several machine learningapproaches have been proposed to forecast and simulate theriver flow time series As a matter of fact the assumption of

the hydrological time series is originated by stochasticitynonlinearity and redundancy Also the underlying mech-anisms of river flow generation are mostly different fromlow- medium- and high-flow periods In other words theextreme events of flows rely upon the outburst of heavystorm of rainfall events Hence developing a new data-in-telligent model characterized by sophistication to forecast allkind of flow patterns satisfactory is still under processmission for the hydrology engineers In this research theapplication of forecasting the monthly river flow (semiaridenvironment) using the hybrid SSA-ELM model is pre-sented Indeed using a univariate modeling is more chal-lenging for the hydrologist as it is fulfilled for the benefit ofthe catchment that lacks the metrological information

(e perdition skills for the conducted modeling werecarried out using two types of performance indicators in-cluding the goodness (eg (r) and (WI)) and the absoluteerror measures (eg (RMSE) and (MAE)) (e mathe-matical description of these performance indicators is

r 1113936

iNi1 Sobsi minus Sobs1113872 1113873 middot S

predi minus Spred1113872 11138731113960 1113961

1113936iNi1 Sobsi minus Sobs1113872 1113873

2middot

1113936iNi1 S

predi minus Spred1113872 1113873

11139691113970

where minus 1le r minus 1

WI 1 minus1113936

iNi1 Sobsi minus S

predi1113872 1113873

2

1113936iNi1 S

predi minus Sobs

11138681113868111386811138681113868

11138681113868111386811138681113868 + Sobsi minus Sobs

11138681113868111386811138681113868

111386811138681113868111386811138681113874 11138752

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

RMSE

1N

1113944

iN

i1Sobsi minus S

predi1113872 1113873

2

11139741113972

MAE 1N

1113944

iN

i1Sobsi minus S

predi

11138681113868111386811138681113868

11138681113868111386811138681113868

(10)

where Sobsi and Spredi are the observed and forecasted values of

river flow Sobs and Spred are the mean values of river flowand N is the number of the observations of the testing phase

For SSA the size of the population is set to 50 while thenumber of iterations is set to 1000 (ese values were rec-ommended and showed superior results when applied forsolving complex optimization problems in [32] For ELMthe common sigmoid function activation function is used[44] However the number of hidden nodes was system-atically determined by experimenting different numbersstarting from 2 hidden nodes up to 10 with a step of 2 nodeseach time

Table 2 tabulates the efficiency results of the hybrid SSA-ELM model achieved based on the testing data for each inputcombination attribute (e results are validated against thepublished results of ELM support vector regression (SVR) andgeneralized regression neural network (GRNN) which wereconducted in [33] on the same case study In addition Table 2also lists the results of random forests (RF) implemented inWeka the popular suite of machine learning software [45]

6 Complexity

Indeed the motivation of the current study is to improvethe performance of the nontuned predictive model ELMthrough complementing with an evolutionary nature-in-spired novel optimization algorithm (ie SSA) Based on the

investigated results an excellent performance in terms oflower RMSE and MAE values and higher r and WI valuesrespectively when forecasting the present river flow datawas observed (e results exhibited a noticeable

Initialize swarm

Calculate the fitness of each particle

ELM network

Assign a vector of inputs weights and

biases to ELM

Update positions of leaders and followers

No

Return best ELM network with

minimum RMSE

Training data

Yes

Evaluate ELM using RMSE

W1 Wnk b0 bk

Maximum number of iterations

reached

Validation data

Figure 6 Description of the proposed SSA-ELM algorithm

Auto

corr

elat

ion

1

05

0

ndash0520181614121086420

Lag (monthly scale)

(a)

1

05

0

ndash0520181614121086420

Lag (monthly scale)

Part

ial a

utoc

orre

latio

ns

(b)

Figure 7 (e statistical correlation process function of the investigated monthly scale river flow

Complexity 7

enhancement in the prediction skills using the proposedSSA-ELM model in comparison with the benchmark ELM-based model (e improved performance with SSA-ELMattributed to the robustness of the novel evolutionary op-timization method SSA linked to the ELM model (isrobustness may have contributed to the internal functionparameter optimization for each input combination

(e Taylor diagram is another modern graphical pre-sentation to diagnose the predictability of the modelingthrough visualizing the errors from the benchmark observedrecord (is diagram graphically summarizes the modelrsquosefficiency in approximating the accuracy to the observeddata (e Taylor diagram of the hybrid SSA-ELM andclassical ELM is illustrated in Figure 8 (e position of eachmodel appeared on the diagram measures the accuracy ofthat model in accordance to the observed river flow Basedon the diagram presentation the SSA-ELM model has ahigher value of correlation compared to the ELMmodel(eSSA-ELM model approximated well with observed andlocated nearer to the actual record (e RMSE between thepredicted and observed river flow was denoted as propor-tional to the distance to the point of the observation (estandard deviation of the predictive model is proportional tothe radial distance from the origin

For detailed inspection for the performance of the SSA-ELM model over the benchmark ELM model Figure 9presents the relative error (RE) distribution indicator inthe form of boxplot Based on the generated graph thehybrid SSA-ELM performed an acceptable improvementwith average RE between minus 10 and + 10 and over 80 of thetesting period(emaximumREmagnitude recorded on the

overestimation is +35 whereas the underestimation aspectis recorded as minus 25(is is much different from the classicalELM that exceeded the minus 40 in the underestimation aspect

Scatter plots for the ELM and SSA-ELM models arepresented in Figures 10 and 11 respectively (e graphicalpresentation followed the determination coefficient values asit is the square value (e proposed hybrid model wasperformed closer to the ideal line between the observed andforecasted values

It is an excellent fashion of evaluating the currentresearch results by comparing with modern establishedresearches with the regards of hybrid evolutionary nature-inspired algorithms integrated with AI-predictive modelsMost recently the applications of metaheuristic techniqueshave shown an extensive and noticeable optimizationmethods for AI predictive models (is is owing to themain advantages such as flexibility give an optimal so-lution and free from gradient decent drawbacks [32] Inthe last five years there are various successful attempts onhybridizing the AI models with evolutionary optimizationalgorithms in the field of hydrology For instance acombination between the adaptive neurofuzzy system(ANFIS) model with the firefly algorithm (FFA) wasconducted for modeling the roller length of the hydraulicjump [46] An integration of the classical multilayer per-ceptron with FFA for pan evaporation process is given in[47] River flow forecasting for tropical environment wasestablished using the ANFIS-FFA model [48] On the sameregion of the case study implemented in this research thewater quality index prediction model was developed basedon the hybridization of the support vector machine model

Table 2 Performance evaluation for ELM and SSA-ELM SVR GRNN and RF models

Input combinations Models r WI RMSE (m3middotsminus 1) MAE (m3 middot sminus 1)

M1

SSA-ELM 0812 0861 80592 66203ELM 081 082 98272 87012SVR 0761 0802 106749 90905

GRNN 0658 0689 127818 113698RF 0372 0531 189778 131966

M2

SSA-ELM 0809 0850 80516 62185ELM 0817 0834 95571 83899SVR 0715 0769 111228 9514

GRNN 0468 054 125187 103551RF 0373 0549 175765 142347

M3

SSA-ELM 0784 0812 86868 62185ELM 0815 0814 104307 92544SVR 0741 0776 11802 10185

GRNN 0496 054 125187 103551RF 0037 033 28085 22527

M4

SSA-ELM 0800 0855 8784 69069ELM 0803 0826 98432 85117SVR 0728 0769 119539 106585

GRNN 0346 0507 133522 114063RF 0127 039 124648 196745

M5

SSA-ELM 0801 0856 87659 67886ELM 0799 0853 87906 71544SVR 0715 0768 124155 108367

GRNN 0248 0416 13535 112606RF 0222 00001 66876 657031

8 Complexity

with the firefly algorithm [49] All the examples exhibitedearlier on the hybrid models demonstrated an excellentperformance over the AI-based models It is evidenced thatthe present study was performed in harmony with a newera of the hybrid intelligent models in the field of surfacehydrology

For the best knowledge of the practical implementationthe proposed hybrid intelligence SSA-ELM model has thepossibility of the online learning potential As reported inFigure 12 the scheme demonstrated the capacity of the onlineforecasting system in which there is a possibility to bepracticed as a floodingwarning system(e online forecastingsystem starts by obtaining the metrological information (eg

river flow velocity (V) cross section area (A) and riverdischarge (Q)) using sensors device(e data information wastransferred via cricket wireless and digitized into a commaseparated value (CSV) format in which it is readable by anysoftware environment At this stage the learning processbased on the historical records is conducted using the SSA-ELM predictive model (e outcome of the predictive modelwas subjected to a specific range of river flow threshold valuesIf the prediction value is within the threshold the loop of thelearning process is repeated using the new recalled historicalinformation Otherwise if the prediction is out of thethreshold range a notification of the alarm system is inte-grated (global system for mobile communications) that can

ELM

SSA-ELM

Observed

00

01

02

03

04

05

06

07

08

09

095

099

0

50

100

150

200

250

0 50 100 150 200 250

Stan

dard

dev

iatio

n

Standard deviation

Taylor diagram

Correlation

Figure 8 Taylor diagram graphical presentation for the ELM and SSA-ELM predictive models

RE

40

30

20

10

0

ndash10

ndash20

ndash30

ndash40

ndash50

RE SSA-ELMRE ELM

Relative error distribution

Figure 9 Relative error distribution percentage of SSA-ELM and ELM models

Complexity 9

y = 0722x + 1487R2 = 0659

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 07228x + 13008R2 = 06553

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 0694x + 13341R2 = 06147

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 0818x + 10463R2 = 06403

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 4

(d)

y = 0811x + 11097R2 = 06422

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 5

(e)

Figure 10 Scatter plot variation between the observed and forecasted river flow using the SSA-ELM model and for the five inputcombinations

10 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

(en the fitness of each particle is calculated using the fol-lowing steps

(i) Assign the values of the input weights and hiddenbiases that a given particle carries to a networkstructure

(ii) Calculate the output weights of the ELM networkusing the PM method based on the training part ofthe dataset

(iii) Calculate the RMSE as given in equation (8) basedon the validation part of the dataset

After performing the fitness evaluation step the particlesare sorted according to their fitness values (e best Lparticles that generate ELM networks with lowest RMSEvalues are nominated as leaders (e rest of the particles inthe swarm are considered as followers(e description of theproposed predictive algorithm is displayed in Figure 6

(e positions of the leaders and followers are updatedaccording to equations ((1) and (3)) respectively (eevaluation and update steps are repeated until a predefinednumber of iterations is reached

(e training procedure of the proposed model isestablished using correlated lags computed via the corre-lations statistical approach (ie autocorrelation and partialautocorrelation functions) to detect appropriate input at-tributes to the targeted attribute (see Figure 7) (is processis accomplished in harmony with major researches of theliterature [42 43] Based on the visualization of the corre-lation statistic five lags time of river flow historical infor-mation (attributes variables) influence the one-step aheadriver flow modeling Hence a combination of five scenariosmodel (M1 M2 M5) architecture is constructed asfollows

Model 1 M1 Q(t) fQ(tminus 1)

Model 2 M1 Q(t) fQ(tminus 1) Q(tminus 2)

Model 3 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3)

Model 4 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4)

Model 5 M1 Q(t) fQ(tminus 1) Q(tminus 2) Q(tminus 3) Q(tminus 4) Q(tminus 5)

(9)

where Q(t) is the target river flow step ldquoone month aheadrdquoand Q(tminus 1) Q(tminus 5) are the lag times up to five monthspreviously Here five input combinations are constructedfor the prediction matrix of the SSA-ELM and ELM pre-dictive models

5 Application Results and Analysis

Over the past couple decades massive attention has beenfocused on hydrological time series modeling particularlyfor river flow process (is is owing to the need of an ac-curate and reliable intelligent model that can be imple-mented in real practice Several machine learningapproaches have been proposed to forecast and simulate theriver flow time series As a matter of fact the assumption of

the hydrological time series is originated by stochasticitynonlinearity and redundancy Also the underlying mech-anisms of river flow generation are mostly different fromlow- medium- and high-flow periods In other words theextreme events of flows rely upon the outburst of heavystorm of rainfall events Hence developing a new data-in-telligent model characterized by sophistication to forecast allkind of flow patterns satisfactory is still under processmission for the hydrology engineers In this research theapplication of forecasting the monthly river flow (semiaridenvironment) using the hybrid SSA-ELM model is pre-sented Indeed using a univariate modeling is more chal-lenging for the hydrologist as it is fulfilled for the benefit ofthe catchment that lacks the metrological information

(e perdition skills for the conducted modeling werecarried out using two types of performance indicators in-cluding the goodness (eg (r) and (WI)) and the absoluteerror measures (eg (RMSE) and (MAE)) (e mathe-matical description of these performance indicators is

r 1113936

iNi1 Sobsi minus Sobs1113872 1113873 middot S

predi minus Spred1113872 11138731113960 1113961

1113936iNi1 Sobsi minus Sobs1113872 1113873

2middot

1113936iNi1 S

predi minus Spred1113872 1113873

11139691113970

where minus 1le r minus 1

WI 1 minus1113936

iNi1 Sobsi minus S

predi1113872 1113873

2

1113936iNi1 S

predi minus Sobs

11138681113868111386811138681113868

11138681113868111386811138681113868 + Sobsi minus Sobs

11138681113868111386811138681113868

111386811138681113868111386811138681113874 11138752

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

RMSE

1N

1113944

iN

i1Sobsi minus S

predi1113872 1113873

2

11139741113972

MAE 1N

1113944

iN

i1Sobsi minus S

predi

11138681113868111386811138681113868

11138681113868111386811138681113868

(10)

where Sobsi and Spredi are the observed and forecasted values of

river flow Sobs and Spred are the mean values of river flowand N is the number of the observations of the testing phase

For SSA the size of the population is set to 50 while thenumber of iterations is set to 1000 (ese values were rec-ommended and showed superior results when applied forsolving complex optimization problems in [32] For ELMthe common sigmoid function activation function is used[44] However the number of hidden nodes was system-atically determined by experimenting different numbersstarting from 2 hidden nodes up to 10 with a step of 2 nodeseach time

Table 2 tabulates the efficiency results of the hybrid SSA-ELM model achieved based on the testing data for each inputcombination attribute (e results are validated against thepublished results of ELM support vector regression (SVR) andgeneralized regression neural network (GRNN) which wereconducted in [33] on the same case study In addition Table 2also lists the results of random forests (RF) implemented inWeka the popular suite of machine learning software [45]

6 Complexity

Indeed the motivation of the current study is to improvethe performance of the nontuned predictive model ELMthrough complementing with an evolutionary nature-in-spired novel optimization algorithm (ie SSA) Based on the

investigated results an excellent performance in terms oflower RMSE and MAE values and higher r and WI valuesrespectively when forecasting the present river flow datawas observed (e results exhibited a noticeable

Initialize swarm

Calculate the fitness of each particle

ELM network

Assign a vector of inputs weights and

biases to ELM

Update positions of leaders and followers

No

Return best ELM network with

minimum RMSE

Training data

Yes

Evaluate ELM using RMSE

W1 Wnk b0 bk

Maximum number of iterations

reached

Validation data

Figure 6 Description of the proposed SSA-ELM algorithm

Auto

corr

elat

ion

1

05

0

ndash0520181614121086420

Lag (monthly scale)

(a)

1

05

0

ndash0520181614121086420

Lag (monthly scale)

Part

ial a

utoc

orre

latio

ns

(b)

Figure 7 (e statistical correlation process function of the investigated monthly scale river flow

Complexity 7

enhancement in the prediction skills using the proposedSSA-ELM model in comparison with the benchmark ELM-based model (e improved performance with SSA-ELMattributed to the robustness of the novel evolutionary op-timization method SSA linked to the ELM model (isrobustness may have contributed to the internal functionparameter optimization for each input combination

(e Taylor diagram is another modern graphical pre-sentation to diagnose the predictability of the modelingthrough visualizing the errors from the benchmark observedrecord (is diagram graphically summarizes the modelrsquosefficiency in approximating the accuracy to the observeddata (e Taylor diagram of the hybrid SSA-ELM andclassical ELM is illustrated in Figure 8 (e position of eachmodel appeared on the diagram measures the accuracy ofthat model in accordance to the observed river flow Basedon the diagram presentation the SSA-ELM model has ahigher value of correlation compared to the ELMmodel(eSSA-ELM model approximated well with observed andlocated nearer to the actual record (e RMSE between thepredicted and observed river flow was denoted as propor-tional to the distance to the point of the observation (estandard deviation of the predictive model is proportional tothe radial distance from the origin

For detailed inspection for the performance of the SSA-ELM model over the benchmark ELM model Figure 9presents the relative error (RE) distribution indicator inthe form of boxplot Based on the generated graph thehybrid SSA-ELM performed an acceptable improvementwith average RE between minus 10 and + 10 and over 80 of thetesting period(emaximumREmagnitude recorded on the

overestimation is +35 whereas the underestimation aspectis recorded as minus 25(is is much different from the classicalELM that exceeded the minus 40 in the underestimation aspect

Scatter plots for the ELM and SSA-ELM models arepresented in Figures 10 and 11 respectively (e graphicalpresentation followed the determination coefficient values asit is the square value (e proposed hybrid model wasperformed closer to the ideal line between the observed andforecasted values

It is an excellent fashion of evaluating the currentresearch results by comparing with modern establishedresearches with the regards of hybrid evolutionary nature-inspired algorithms integrated with AI-predictive modelsMost recently the applications of metaheuristic techniqueshave shown an extensive and noticeable optimizationmethods for AI predictive models (is is owing to themain advantages such as flexibility give an optimal so-lution and free from gradient decent drawbacks [32] Inthe last five years there are various successful attempts onhybridizing the AI models with evolutionary optimizationalgorithms in the field of hydrology For instance acombination between the adaptive neurofuzzy system(ANFIS) model with the firefly algorithm (FFA) wasconducted for modeling the roller length of the hydraulicjump [46] An integration of the classical multilayer per-ceptron with FFA for pan evaporation process is given in[47] River flow forecasting for tropical environment wasestablished using the ANFIS-FFA model [48] On the sameregion of the case study implemented in this research thewater quality index prediction model was developed basedon the hybridization of the support vector machine model

Table 2 Performance evaluation for ELM and SSA-ELM SVR GRNN and RF models

Input combinations Models r WI RMSE (m3middotsminus 1) MAE (m3 middot sminus 1)

M1

SSA-ELM 0812 0861 80592 66203ELM 081 082 98272 87012SVR 0761 0802 106749 90905

GRNN 0658 0689 127818 113698RF 0372 0531 189778 131966

M2

SSA-ELM 0809 0850 80516 62185ELM 0817 0834 95571 83899SVR 0715 0769 111228 9514

GRNN 0468 054 125187 103551RF 0373 0549 175765 142347

M3

SSA-ELM 0784 0812 86868 62185ELM 0815 0814 104307 92544SVR 0741 0776 11802 10185

GRNN 0496 054 125187 103551RF 0037 033 28085 22527

M4

SSA-ELM 0800 0855 8784 69069ELM 0803 0826 98432 85117SVR 0728 0769 119539 106585

GRNN 0346 0507 133522 114063RF 0127 039 124648 196745

M5

SSA-ELM 0801 0856 87659 67886ELM 0799 0853 87906 71544SVR 0715 0768 124155 108367

GRNN 0248 0416 13535 112606RF 0222 00001 66876 657031

8 Complexity

with the firefly algorithm [49] All the examples exhibitedearlier on the hybrid models demonstrated an excellentperformance over the AI-based models It is evidenced thatthe present study was performed in harmony with a newera of the hybrid intelligent models in the field of surfacehydrology

For the best knowledge of the practical implementationthe proposed hybrid intelligence SSA-ELM model has thepossibility of the online learning potential As reported inFigure 12 the scheme demonstrated the capacity of the onlineforecasting system in which there is a possibility to bepracticed as a floodingwarning system(e online forecastingsystem starts by obtaining the metrological information (eg

river flow velocity (V) cross section area (A) and riverdischarge (Q)) using sensors device(e data information wastransferred via cricket wireless and digitized into a commaseparated value (CSV) format in which it is readable by anysoftware environment At this stage the learning processbased on the historical records is conducted using the SSA-ELM predictive model (e outcome of the predictive modelwas subjected to a specific range of river flow threshold valuesIf the prediction value is within the threshold the loop of thelearning process is repeated using the new recalled historicalinformation Otherwise if the prediction is out of thethreshold range a notification of the alarm system is inte-grated (global system for mobile communications) that can

ELM

SSA-ELM

Observed

00

01

02

03

04

05

06

07

08

09

095

099

0

50

100

150

200

250

0 50 100 150 200 250

Stan

dard

dev

iatio

n

Standard deviation

Taylor diagram

Correlation

Figure 8 Taylor diagram graphical presentation for the ELM and SSA-ELM predictive models

RE

40

30

20

10

0

ndash10

ndash20

ndash30

ndash40

ndash50

RE SSA-ELMRE ELM

Relative error distribution

Figure 9 Relative error distribution percentage of SSA-ELM and ELM models

Complexity 9

y = 0722x + 1487R2 = 0659

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 07228x + 13008R2 = 06553

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 0694x + 13341R2 = 06147

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 0818x + 10463R2 = 06403

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 4

(d)

y = 0811x + 11097R2 = 06422

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 5

(e)

Figure 10 Scatter plot variation between the observed and forecasted river flow using the SSA-ELM model and for the five inputcombinations

10 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

Indeed the motivation of the current study is to improvethe performance of the nontuned predictive model ELMthrough complementing with an evolutionary nature-in-spired novel optimization algorithm (ie SSA) Based on the

investigated results an excellent performance in terms oflower RMSE and MAE values and higher r and WI valuesrespectively when forecasting the present river flow datawas observed (e results exhibited a noticeable

Initialize swarm

Calculate the fitness of each particle

ELM network

Assign a vector of inputs weights and

biases to ELM

Update positions of leaders and followers

No

Return best ELM network with

minimum RMSE

Training data

Yes

Evaluate ELM using RMSE

W1 Wnk b0 bk

Maximum number of iterations

reached

Validation data

Figure 6 Description of the proposed SSA-ELM algorithm

Auto

corr

elat

ion

1

05

0

ndash0520181614121086420

Lag (monthly scale)

(a)

1

05

0

ndash0520181614121086420

Lag (monthly scale)

Part

ial a

utoc

orre

latio

ns

(b)

Figure 7 (e statistical correlation process function of the investigated monthly scale river flow

Complexity 7

enhancement in the prediction skills using the proposedSSA-ELM model in comparison with the benchmark ELM-based model (e improved performance with SSA-ELMattributed to the robustness of the novel evolutionary op-timization method SSA linked to the ELM model (isrobustness may have contributed to the internal functionparameter optimization for each input combination

(e Taylor diagram is another modern graphical pre-sentation to diagnose the predictability of the modelingthrough visualizing the errors from the benchmark observedrecord (is diagram graphically summarizes the modelrsquosefficiency in approximating the accuracy to the observeddata (e Taylor diagram of the hybrid SSA-ELM andclassical ELM is illustrated in Figure 8 (e position of eachmodel appeared on the diagram measures the accuracy ofthat model in accordance to the observed river flow Basedon the diagram presentation the SSA-ELM model has ahigher value of correlation compared to the ELMmodel(eSSA-ELM model approximated well with observed andlocated nearer to the actual record (e RMSE between thepredicted and observed river flow was denoted as propor-tional to the distance to the point of the observation (estandard deviation of the predictive model is proportional tothe radial distance from the origin

For detailed inspection for the performance of the SSA-ELM model over the benchmark ELM model Figure 9presents the relative error (RE) distribution indicator inthe form of boxplot Based on the generated graph thehybrid SSA-ELM performed an acceptable improvementwith average RE between minus 10 and + 10 and over 80 of thetesting period(emaximumREmagnitude recorded on the

overestimation is +35 whereas the underestimation aspectis recorded as minus 25(is is much different from the classicalELM that exceeded the minus 40 in the underestimation aspect

Scatter plots for the ELM and SSA-ELM models arepresented in Figures 10 and 11 respectively (e graphicalpresentation followed the determination coefficient values asit is the square value (e proposed hybrid model wasperformed closer to the ideal line between the observed andforecasted values

It is an excellent fashion of evaluating the currentresearch results by comparing with modern establishedresearches with the regards of hybrid evolutionary nature-inspired algorithms integrated with AI-predictive modelsMost recently the applications of metaheuristic techniqueshave shown an extensive and noticeable optimizationmethods for AI predictive models (is is owing to themain advantages such as flexibility give an optimal so-lution and free from gradient decent drawbacks [32] Inthe last five years there are various successful attempts onhybridizing the AI models with evolutionary optimizationalgorithms in the field of hydrology For instance acombination between the adaptive neurofuzzy system(ANFIS) model with the firefly algorithm (FFA) wasconducted for modeling the roller length of the hydraulicjump [46] An integration of the classical multilayer per-ceptron with FFA for pan evaporation process is given in[47] River flow forecasting for tropical environment wasestablished using the ANFIS-FFA model [48] On the sameregion of the case study implemented in this research thewater quality index prediction model was developed basedon the hybridization of the support vector machine model

Table 2 Performance evaluation for ELM and SSA-ELM SVR GRNN and RF models

Input combinations Models r WI RMSE (m3middotsminus 1) MAE (m3 middot sminus 1)

M1

SSA-ELM 0812 0861 80592 66203ELM 081 082 98272 87012SVR 0761 0802 106749 90905

GRNN 0658 0689 127818 113698RF 0372 0531 189778 131966

M2

SSA-ELM 0809 0850 80516 62185ELM 0817 0834 95571 83899SVR 0715 0769 111228 9514

GRNN 0468 054 125187 103551RF 0373 0549 175765 142347

M3

SSA-ELM 0784 0812 86868 62185ELM 0815 0814 104307 92544SVR 0741 0776 11802 10185

GRNN 0496 054 125187 103551RF 0037 033 28085 22527

M4

SSA-ELM 0800 0855 8784 69069ELM 0803 0826 98432 85117SVR 0728 0769 119539 106585

GRNN 0346 0507 133522 114063RF 0127 039 124648 196745

M5

SSA-ELM 0801 0856 87659 67886ELM 0799 0853 87906 71544SVR 0715 0768 124155 108367

GRNN 0248 0416 13535 112606RF 0222 00001 66876 657031

8 Complexity

with the firefly algorithm [49] All the examples exhibitedearlier on the hybrid models demonstrated an excellentperformance over the AI-based models It is evidenced thatthe present study was performed in harmony with a newera of the hybrid intelligent models in the field of surfacehydrology

For the best knowledge of the practical implementationthe proposed hybrid intelligence SSA-ELM model has thepossibility of the online learning potential As reported inFigure 12 the scheme demonstrated the capacity of the onlineforecasting system in which there is a possibility to bepracticed as a floodingwarning system(e online forecastingsystem starts by obtaining the metrological information (eg

river flow velocity (V) cross section area (A) and riverdischarge (Q)) using sensors device(e data information wastransferred via cricket wireless and digitized into a commaseparated value (CSV) format in which it is readable by anysoftware environment At this stage the learning processbased on the historical records is conducted using the SSA-ELM predictive model (e outcome of the predictive modelwas subjected to a specific range of river flow threshold valuesIf the prediction value is within the threshold the loop of thelearning process is repeated using the new recalled historicalinformation Otherwise if the prediction is out of thethreshold range a notification of the alarm system is inte-grated (global system for mobile communications) that can

ELM

SSA-ELM

Observed

00

01

02

03

04

05

06

07

08

09

095

099

0

50

100

150

200

250

0 50 100 150 200 250

Stan

dard

dev

iatio

n

Standard deviation

Taylor diagram

Correlation

Figure 8 Taylor diagram graphical presentation for the ELM and SSA-ELM predictive models

RE

40

30

20

10

0

ndash10

ndash20

ndash30

ndash40

ndash50

RE SSA-ELMRE ELM

Relative error distribution

Figure 9 Relative error distribution percentage of SSA-ELM and ELM models

Complexity 9

y = 0722x + 1487R2 = 0659

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 07228x + 13008R2 = 06553

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 0694x + 13341R2 = 06147

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 0818x + 10463R2 = 06403

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 4

(d)

y = 0811x + 11097R2 = 06422

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 5

(e)

Figure 10 Scatter plot variation between the observed and forecasted river flow using the SSA-ELM model and for the five inputcombinations

10 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

enhancement in the prediction skills using the proposedSSA-ELM model in comparison with the benchmark ELM-based model (e improved performance with SSA-ELMattributed to the robustness of the novel evolutionary op-timization method SSA linked to the ELM model (isrobustness may have contributed to the internal functionparameter optimization for each input combination

(e Taylor diagram is another modern graphical pre-sentation to diagnose the predictability of the modelingthrough visualizing the errors from the benchmark observedrecord (is diagram graphically summarizes the modelrsquosefficiency in approximating the accuracy to the observeddata (e Taylor diagram of the hybrid SSA-ELM andclassical ELM is illustrated in Figure 8 (e position of eachmodel appeared on the diagram measures the accuracy ofthat model in accordance to the observed river flow Basedon the diagram presentation the SSA-ELM model has ahigher value of correlation compared to the ELMmodel(eSSA-ELM model approximated well with observed andlocated nearer to the actual record (e RMSE between thepredicted and observed river flow was denoted as propor-tional to the distance to the point of the observation (estandard deviation of the predictive model is proportional tothe radial distance from the origin

For detailed inspection for the performance of the SSA-ELM model over the benchmark ELM model Figure 9presents the relative error (RE) distribution indicator inthe form of boxplot Based on the generated graph thehybrid SSA-ELM performed an acceptable improvementwith average RE between minus 10 and + 10 and over 80 of thetesting period(emaximumREmagnitude recorded on the

overestimation is +35 whereas the underestimation aspectis recorded as minus 25(is is much different from the classicalELM that exceeded the minus 40 in the underestimation aspect

Scatter plots for the ELM and SSA-ELM models arepresented in Figures 10 and 11 respectively (e graphicalpresentation followed the determination coefficient values asit is the square value (e proposed hybrid model wasperformed closer to the ideal line between the observed andforecasted values

It is an excellent fashion of evaluating the currentresearch results by comparing with modern establishedresearches with the regards of hybrid evolutionary nature-inspired algorithms integrated with AI-predictive modelsMost recently the applications of metaheuristic techniqueshave shown an extensive and noticeable optimizationmethods for AI predictive models (is is owing to themain advantages such as flexibility give an optimal so-lution and free from gradient decent drawbacks [32] Inthe last five years there are various successful attempts onhybridizing the AI models with evolutionary optimizationalgorithms in the field of hydrology For instance acombination between the adaptive neurofuzzy system(ANFIS) model with the firefly algorithm (FFA) wasconducted for modeling the roller length of the hydraulicjump [46] An integration of the classical multilayer per-ceptron with FFA for pan evaporation process is given in[47] River flow forecasting for tropical environment wasestablished using the ANFIS-FFA model [48] On the sameregion of the case study implemented in this research thewater quality index prediction model was developed basedon the hybridization of the support vector machine model

Table 2 Performance evaluation for ELM and SSA-ELM SVR GRNN and RF models

Input combinations Models r WI RMSE (m3middotsminus 1) MAE (m3 middot sminus 1)

M1

SSA-ELM 0812 0861 80592 66203ELM 081 082 98272 87012SVR 0761 0802 106749 90905

GRNN 0658 0689 127818 113698RF 0372 0531 189778 131966

M2

SSA-ELM 0809 0850 80516 62185ELM 0817 0834 95571 83899SVR 0715 0769 111228 9514

GRNN 0468 054 125187 103551RF 0373 0549 175765 142347

M3

SSA-ELM 0784 0812 86868 62185ELM 0815 0814 104307 92544SVR 0741 0776 11802 10185

GRNN 0496 054 125187 103551RF 0037 033 28085 22527

M4

SSA-ELM 0800 0855 8784 69069ELM 0803 0826 98432 85117SVR 0728 0769 119539 106585

GRNN 0346 0507 133522 114063RF 0127 039 124648 196745

M5

SSA-ELM 0801 0856 87659 67886ELM 0799 0853 87906 71544SVR 0715 0768 124155 108367

GRNN 0248 0416 13535 112606RF 0222 00001 66876 657031

8 Complexity

with the firefly algorithm [49] All the examples exhibitedearlier on the hybrid models demonstrated an excellentperformance over the AI-based models It is evidenced thatthe present study was performed in harmony with a newera of the hybrid intelligent models in the field of surfacehydrology

For the best knowledge of the practical implementationthe proposed hybrid intelligence SSA-ELM model has thepossibility of the online learning potential As reported inFigure 12 the scheme demonstrated the capacity of the onlineforecasting system in which there is a possibility to bepracticed as a floodingwarning system(e online forecastingsystem starts by obtaining the metrological information (eg

river flow velocity (V) cross section area (A) and riverdischarge (Q)) using sensors device(e data information wastransferred via cricket wireless and digitized into a commaseparated value (CSV) format in which it is readable by anysoftware environment At this stage the learning processbased on the historical records is conducted using the SSA-ELM predictive model (e outcome of the predictive modelwas subjected to a specific range of river flow threshold valuesIf the prediction value is within the threshold the loop of thelearning process is repeated using the new recalled historicalinformation Otherwise if the prediction is out of thethreshold range a notification of the alarm system is inte-grated (global system for mobile communications) that can

ELM

SSA-ELM

Observed

00

01

02

03

04

05

06

07

08

09

095

099

0

50

100

150

200

250

0 50 100 150 200 250

Stan

dard

dev

iatio

n

Standard deviation

Taylor diagram

Correlation

Figure 8 Taylor diagram graphical presentation for the ELM and SSA-ELM predictive models

RE

40

30

20

10

0

ndash10

ndash20

ndash30

ndash40

ndash50

RE SSA-ELMRE ELM

Relative error distribution

Figure 9 Relative error distribution percentage of SSA-ELM and ELM models

Complexity 9

y = 0722x + 1487R2 = 0659

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 07228x + 13008R2 = 06553

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 0694x + 13341R2 = 06147

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 0818x + 10463R2 = 06403

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 4

(d)

y = 0811x + 11097R2 = 06422

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 5

(e)

Figure 10 Scatter plot variation between the observed and forecasted river flow using the SSA-ELM model and for the five inputcombinations

10 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

with the firefly algorithm [49] All the examples exhibitedearlier on the hybrid models demonstrated an excellentperformance over the AI-based models It is evidenced thatthe present study was performed in harmony with a newera of the hybrid intelligent models in the field of surfacehydrology

For the best knowledge of the practical implementationthe proposed hybrid intelligence SSA-ELM model has thepossibility of the online learning potential As reported inFigure 12 the scheme demonstrated the capacity of the onlineforecasting system in which there is a possibility to bepracticed as a floodingwarning system(e online forecastingsystem starts by obtaining the metrological information (eg

river flow velocity (V) cross section area (A) and riverdischarge (Q)) using sensors device(e data information wastransferred via cricket wireless and digitized into a commaseparated value (CSV) format in which it is readable by anysoftware environment At this stage the learning processbased on the historical records is conducted using the SSA-ELM predictive model (e outcome of the predictive modelwas subjected to a specific range of river flow threshold valuesIf the prediction value is within the threshold the loop of thelearning process is repeated using the new recalled historicalinformation Otherwise if the prediction is out of thethreshold range a notification of the alarm system is inte-grated (global system for mobile communications) that can

ELM

SSA-ELM

Observed

00

01

02

03

04

05

06

07

08

09

095

099

0

50

100

150

200

250

0 50 100 150 200 250

Stan

dard

dev

iatio

n

Standard deviation

Taylor diagram

Correlation

Figure 8 Taylor diagram graphical presentation for the ELM and SSA-ELM predictive models

RE

40

30

20

10

0

ndash10

ndash20

ndash30

ndash40

ndash50

RE SSA-ELMRE ELM

Relative error distribution

Figure 9 Relative error distribution percentage of SSA-ELM and ELM models

Complexity 9

y = 0722x + 1487R2 = 0659

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 07228x + 13008R2 = 06553

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 0694x + 13341R2 = 06147

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 0818x + 10463R2 = 06403

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 4

(d)

y = 0811x + 11097R2 = 06422

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 5

(e)

Figure 10 Scatter plot variation between the observed and forecasted river flow using the SSA-ELM model and for the five inputcombinations

10 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

y = 0722x + 1487R2 = 0659

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 07228x + 13008R2 = 06553

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 0694x + 13341R2 = 06147

300

400

500

600

700

800

900

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 0818x + 10463R2 = 06403

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 4

(d)

y = 0811x + 11097R2 = 06422

300

400

500

600

700

800

900

1000

300 500 700 900 1100

SSA

-ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 5

(e)

Figure 10 Scatter plot variation between the observed and forecasted river flow using the SSA-ELM model and for the five inputcombinations

10 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

y = 06072x + 25092R2 = 06561

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 1

(a)

y = 06516x + 22669R2 = 06671

300

400

500

600

700

800

900

1000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 2

(b)

y = 06237x + 25387R2 = 06646

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Obsevred (m3middotsndash1)

Lag time 3

(c)

y = 06442x + 23159Rsup2 = 0644

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 4

(d)

y = 07559x + 14695R2 = 06385

30000

40000

50000

60000

70000

80000

90000

100000

300 500 700 900 1100

ELM

(m3 middotsndash1

)

Observed (m3middotsndash1)

Lag time 5

(e)

Figure 11 Scatter plot variation between the observed and forecasted river flow using the ELMmodel and for the five input combinations

Complexity 11

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

deliver a message of alarm to the nearby civilian for preparingan emergency evacuation (is merit can contribute re-markably to the water resources engineering river man-agement and monitoring perspective

6 Conclusions

In this research a novel hybrid intelligent predictivemodel wasdeveloped to be implemented on a regression hydrologicalproblem (ie river flow forecasting) A new nature-inspiredoptimization algorithm called Salp Swarm Algorithm wasintegrated with the modern AI model (ie ELM model) tobuild a robust intelligent predictivemodel(e basic concept ofthe introduced SSA is inspired from the swarming behaviour ofthe salp chain(e developed SSA-ELMmodel was applied formonthly time scale river flow data located in semiarid envi-ronment in the Iraq region Twenty years over the period(1991ndash2010) was used to design the training and testing phasesfollowing the published research [33] Several input combi-nation attributes were investigated in accordance with themostcorrelated lag times (e hybrid SSA-ELM model exhibited anexcellent level of accuracies in comparison with the ELM-basedmodel In numerical evaluation the SSA-ELM model dem-onstrated an acceptable level of augmentation with regards tothe absolute metrics (84 and 131 for RMSE and MAErespectively) On the other aspect the results verified that thelag time as an input attribute is very significant on themodelingperformance of the predictive model In conclusion the de-veloped model outcomes enthuse for proposing an onlineforecasting system that can be applied as a real-time moni-toring river engineering practice For future work furtherresearch can be conducted based on investigating the efficiencyof the proposed model by incorporating several hydrological

processes such as rainfall humidity wind speed and others inmodeling the target output river flow [50] In additioninspecting longer data span of historical data information ishighly recommended to be devoted as it is a limitation of thecurrent research On the algorithmic level the proposed modelcan be extended in a way that other parameters in the ELMnetwork can be optimized and such parameters include thenumber of hidden nodes in the hidden layer the transferfunctions and the activation functions [51ndash53]

Data Availability

(e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

(e authors have no conflicts of interest to declare

References

[1] J Patskoski and A Sankarasubramanian ldquoReducing uncer-tainty in stochastic streamflow generation and reservoir sizingby combining observed reconstructed and projectedstreamflowrdquo Stochastic Environmental Research and RiskAssessment vol 32 no 4 pp 1065ndash1083 2018

[2] F Fathian A Fakheri-Fard T Ouarda Y Dinpashoh andS S M Nadoushani ldquoMultiple streamflow time series modelingusing VARndashMGARCH approachrdquo Stochastic EnvironmentalResearch and Risk Assessment vol 32 no 2 pp 407ndash425 2019

[3] M Azmat M Choi T-W Kim and U W Liaqat ldquoHy-drological modeling to simulate streamflow under changingclimate in a scarcely gauged cryosphere catchmentrdquo Envi-ronmental Earth Sciences vol 75 no 3 pp 1ndash16 2016

Sensers-V A Q DigitizationCricket wireless CSV form

SSA-ELM predictive modelThreshold

Updated system

Prediction decision

Within range

Over range

Global system for mobile communications

Predictivemodeling

Figure 12 A proposition for online forecasting system based on the proposed methodology

12 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

[4] K Stahl R D Moore J M Shea D Hutchinson andA J Cannon ldquoCoupled modelling of glacier and streamflowresponse to future climate scenariosrdquo Water Resources Re-search vol 44 2008

[5] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2010

[6] R Modarres and S S Eslamian ldquoStreamflow time seriesmodeling of Zayandehrud Riverrdquo Iranian Journal of Scienceand Technology Transaction B Engineering vol 30 pp 567ndash570 2006

[7] I Masih S Maskey S Uhlenbrook and V SmakhtinldquoAssessing the impact of areal precipitation input onstreamflow simulations using the SWAT Model1rdquo JAWRAJournal of the American Water Resources Association vol 47no 1 pp 179ndash195 2011

[8] K Rahman C Maringanti M Beniston F WidmerK Abbaspour and A Lehmann ldquoStreamflow modeling in ahighly managed mountainous glacier watershed using SWATthe upper Rhone River watershed case in SwitzerlandrdquoWaterResources Management vol 27 no 2 pp 323ndash339 2013

[9] Z M Yaseen A El-shafie O Jaafar H A Afan andK N Sayl ldquoArtificial intelligence based models for stream-flow forecasting 2000ndash2015rdquo Journal of Hydrology vol 530pp 829ndash844 2015

[10] W Hu L Yan K Liu and H Wang ldquoA short-term trafficflow forecasting method based on the hybrid PSO-SVRrdquoNeural Processing Letters vol 43 no 1 pp 155ndash172 2016

[11] Z M Yaseen O Kisi and V Demir ldquoEnhancing long-termstreamflow forecasting and predicting using periodicity datacomponent application of artificial intelligencerdquo Water Re-sources Management vol 30 no 12 pp 4125ndash4151 2016

[12] A Danandeh Mehr E Kahya F Bagheri and E DeliktasldquoSuccessive-station monthly streamflow prediction usingneuro-wavelet techniquerdquo Earth Science Informatics vol 7no 4 pp 217ndash229 2013

[13] R C Deo andM Sahin ldquoAn extreme learning machine modelfor the simulation of monthly mean streamflow water level ineastern Queenslandrdquo Environmental Monitoring and As-sessment vol 188 no 2 2016

[14] Z M Yaseen S O Sulaiman R C Deo and K-W Chau ldquoAnenhanced extreme learning machine model for river flowforecasting state-of-the-art practical applications in waterresource engineering area and future research directionrdquoJournal of Hydrology vol 569 pp 387ndash408 2019

[15] H Jiang and Y Dong ldquoA novel model based on square rootelastic net and artificial neural network for forecasting globalsolar radiationrdquo Complexity vol 2018 Article ID 813519319 pages 2018

[16] B B Sahoo R Jha A Singh and D Kumar ldquoApplication ofsupport vector regression for modeling low flow time seriesrdquoKSCE Journal of Civil Engineering vol 23 no 2 pp 923ndash9342019

[17] A Mohanta K Patra and B Sahoo ldquoAnticipate Manningrsquoscoefficient in meandering compound channelsrdquo Hydrologyvol 5 no 3 p 47 2018

[18] B B Sahoo R Jha A Singh and D Kumar ldquoLong short-termmemory (LSTM) recurrent neural network for low-flowhydrological time series forecastingrdquo Acta Geophysica vol 67no 5 pp 1471ndash1481 2019

[19] J L Salmeron M B Correia and P R Palos-SanchezldquoComplexity in forecasting and predictive modelsrdquo Com-plexity vol 2019 Article ID 8160659 3 pages 2019

[20] G-B Huang Q-Y Zhu and C-K Siew ldquoExtreme learningmachine theory and applicationsrdquo Neurocomputing vol 70no 1ndash3 pp 489ndash501 2006

[21] H Siqueira L Boccato R Attux and C Lyra ldquoUnorganizedmachines for seasonal streamflow series forecastingrdquo Inter-national Journal of Neural Systems vol 24 no 3 Article ID1430009 2014

[22] A R Lima A J Cannon andWW Hsieh ldquoForecasting dailystreamflow using online sequential extreme learning ma-chinesrdquo Journal of Hydrology vol 537 pp 431ndash443 2016

[23] B Yadav S Ch S Mathur and J Adamowski ldquoDischargeforecasting using an online sequential extreme learningmachine (OS-ELM) model a case study in Neckar RiverGermanyrdquo Measurement vol 92 pp 433ndash445 2016

[24] M Rezaie-Balf and O Kisi ldquoNew formulation for forecastingstreamflow evolutionary polynomial regression vs extremelearning machinerdquo Hydrology Research vol 49 no 3pp 939ndash953 2017

[25] Z M Yaseen M F Allawi A A Yousif O JaafarF M Hamzah and A El-Shafie ldquoNon-tuned machinelearning approach for hydrological time series forecastingrdquoNeural Computing and Applications vol 30 no 5pp 1479ndash1491 2018

[26] A R Lima A J Cannon and W W Hsieh ldquoNonlinearregression in environmental sciences using extreme learningmachines a comparative evaluationrdquo Environmental Mod-elling amp Software vol 73 pp 175ndash188 2015

[27] S Heddam and O Kisi ldquoExtreme learning machines a newapproach for modeling dissolved oxygen (DO) concentrationwith and without water quality variables as predictorsrdquo En-vironmental Science and Pollution Research vol 24 no 20pp 16702ndash16724 2017

[28] S Heddam ldquoUse of optimally pruned extreme learningmachine (OP-ELM) in forecasting dissolved oxygen con-centration (DO) several hours in advance a case study fromthe Lamath River Oregon USArdquo Environmental Processesvol 3 no 4 pp 909ndash937 2016

[29] Z M Yaseen S M Awadh A Sharafati and S ShahidldquoComplementary data-intelligence model for river flowsimulationrdquo Journal of Hydrology vol 567 pp 180ndash190 2018

[30] P Mohapatra K Nath Das and S Roy ldquoA modified com-petitive swarm optimizer for large scale optimization prob-lemsrdquo Applied Soft Computing vol 59 pp 340ndash362 2017

[31] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[32] S Mirjalili A H Gandomi S Z Mirjalili S Saremi H Farisand S M Mirjalili ldquoSalp swarm algorithm a bio-inspiredoptimizer for engineering design problemsrdquo Advances inEngineering Software vol 114 pp 163ndash191 2017

[33] Z M Yaseen O Jaafar R C Deo et al ldquoStream-flowforecasting using extreme learning machines a case study in asemi-arid region in Iraqrdquo Journal of Hydrology vol 542pp 603ndash614 2016

[34] G Huang G-B Huang S Song and K You ldquoTrends inextreme learning machines a reviewrdquo Neural Networksvol 61 pp 32ndash48 2015

[35] H Sanikhani R C Deo Z M Yaseen O Eray and O KisildquoNon-tuned data intelligent model for soil temperature es-timation a new approachrdquo Geoderma vol 330 pp 52ndash642018

[36] M Eshtay H Faris and N Obeid ldquoImproving extremelearning machine by competitive swarm optimization and its

Complexity 13

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity

application for medical diagnosis problemsrdquo Expert Systemswith Applications vol 104 pp 134ndash152 2018

[37] N Al-ansari M Abdellatif M Ezeelden S S Ali andS Knutsson ldquoClimate change and future long-term trends ofrainfall at north-east of Iraqrdquo Journal of Civil Engineering andArchitecture vol 8 no 6 pp 790ndash805 2014

[38] D Bozkurt O Sen and S Hagemann ldquoProjected river dis-charge in the Euphrates-Tigris basin from a hydrologicaldischarge model forced with RCM and GCM outputsrdquo Cli-mate Research vol 62 no 2 pp 131ndash147 2015

[39] N Al-Ansari A A Ali and S Knutsson ldquoPresent conditionsand future challenges of water resources problems in IraqrdquoJournal of Water Resource and Protection vol 6 no 12pp 1066ndash1098 2014

[40] S A Salman S Shahid T Ismail N B A Rahman X Wangand E-S Chung ldquoUnidirectional trends in daily rainfallextremes of Iraqrdquo eoretical and Applied Climatologyvol 134 no 3-4 pp 1165ndash1177 2018

[41] D K Saleh Stream Gage Descriptions and Streamflow Sta-tistics for Sites in the Tigris River and Euphrates River BasinsIraq US Department of the Interior US Geological SurveyReston VA USA 2010

[42] K P Sudheer A K Gosain D Mohana Rangan andS M Saheb ldquoModelling evaporation using an artificial neuralnetwork algorithmrdquo Hydrological Processes vol 16 no 16pp 3189ndash3202 2002

[43] O Kisi ldquoMulti-layer perceptrons with Levenberg-Marquardttraining algorithm for suspended sediment concentrationprediction and estimationrdquo Hydrological Sciences Journalvol 49 pp 1025ndash1040 2004

[44] A A H Alwanas A A Al-Musawi S Q Salih H Tao M Aliand Z M Yaseen ldquoLoad-carrying capacity and mode failuresimulation of beam-column joint connection application ofself-tuning machine learning modelrdquo Engineering Structuresvol 194 pp 220ndash229 2019

[45] M Hall E Frank G Holmes B Pfahringer P Reutemannand I H Witten ldquo(e WEKA data mining softwarerdquo ACMSIGKDD Explorations Newsletter vol 11 no 1 p 10 2009

[46] H Azimi H Bonakdari I Ebtehaj and D G Michelson ldquoAcombined adaptive neuro-fuzzy inference system-firefly al-gorithm model for predicting the roller length of a hydraulicjump on a rough channel bedrdquo Neural Computing and Ap-plications vol 29 no 6 pp 249ndash258 2016

[47] M A Ghorbani R C Deo V Karimi Z M Yaseen andO Terzi ldquoImplementation of a hybrid MLP-FFA model forwater level prediction of Lake Egirdir Turkeyrdquo StochasticEnvironmental Research and Risk Assessment vol 32 no 6pp 1683ndash1697 2017

[48] Z M Yaseen I Ebtehaj H Bonakdari et al ldquoNovel approachfor streamflow forecasting using a hybrid ANFIS-FFAmodelrdquoJournal of Hydrology vol 554 pp 263ndash276 2017

[49] J Li H Ali A Samer S Hasan et al ldquoHybrid soft computingapproach for determining water quality indicator EuphratesRiverrdquo Neural Computing and Applications vol 31 no 3pp 827ndash837 2017

[50] L Diop A Bodian K Djaman et al ldquo(e influence of cli-matic inputs on stream-flow pattern forecasting case study ofUpper Senegal Riverrdquo Environmental Earth Sciences vol 77no 5 p 182 2018

[51] S Q Salih and A A Alsewari ldquoSolving large-scale problemsusing multi-swarm particle swarm approachrdquo InternationalJournal of Engineering amp Technology vol 7 no 3 pp 1725ndash1729 2018

[52] S Q Salih and A A Alsewari ldquoA new algorithm for normaland large-scale optimization problems nomadic people op-timizerrdquo Neural Computing and Applicationspp 1ndash28 2019In press

[53] H A Abdulwahab A Noraziah A A Alsewari andS Q Salih ldquoAn enhanced version of black hole algorithm vialevy flight for optimization and data clustering problemsrdquoIEEE Access vol 7 pp 142085ndash142096 2019

14 Complexity