01_5-9-2

9
Performance study of solar-assisted air-conditioning system provided with storage tanks using artificial neural networks S. Rosiek, F.J. Batlles* Dpto. Fı´sica Aplicada, Universidad de Almerı´a, Ctra. Sacramento s/n, La Canada de San Urbano, 04120 Almerı´a, Spain article info Article history: Received 3 December 2009 Received in revised form 29 March 2011 Accepted 2 May 2011 Available online 13 May 2011 Keywords: Thermal storage Absorption system Neural network abstract This study presents the performance of solar-assisted air-conditioning system provided with two storage tanks installed in the Solar Energy Research Center (CIESOL) building. The system consists mainly of solar collectors’ array, a hot-water driven absorption chiller, a cooling tower, two hot storage tanks, an auxiliary heater as well as two cold storage tanks. The hot storage tank circuit was further investigated. In first step, we have carried out the study about the necessity of the integration of hot water storage tanks to solar system. Subsequently, the unique Artificial Neural Network (ANN) model with the lowest number of input variables has been proposed with the main purpose to predict the coef- ficient of performance and the cooling capacity of the absorption chiller. The configuration 5-9-2 (5 inputs, 9 hidden and 2 output neurons) was found to be the optimal topology. The results demonstrate proper ANN’s predictions with a Root Mean Square Error (RMSE) of less than 0.70% and practically null deviation, which can be considered very satisfactory. ª 2011 Elsevier Ltd and IIR. All rights reserved. Etude sur la performance d’un syste ` me de conditionnement d’air fonctionnant en partie gra ˆce a ` l’e ´ nergie solaire et muni de bacs d’accumulation employant des re ´ seaux neuronaux artificiels Mots cle ´s : accumulation thermique ; syste `me a ` absorption ; re ´ seau neuronal 1. Introduction Solar air conditioning is an emerging market with a huge growth potential. Peak cooling demand in summer is associ- ated with high availability of solar radiation, which offers an excellent opportunity to exploit solar energy with heat-driven cooling machines. For low power cooling systems, commer- cial technologies are available on a limited basis. However, a strong attention is put on research on other applications including photovoltaic-operated refrigeration cycles and solar mechanical refrigeration (Balaras et al., 2006). Recently research has been devoted to improving the main * Corresponding author. Tel.: þ34 950 015914; fax: þ34 950 015477. E-mail address: [email protected] (F.J. Batlles). www.iifiir.org available at www.sciencedirect.com journal homepage: www.elsevier.com/locate/ijrefrig international journal of refrigeration 34 (2011) 1446 e1454 0140-7007/$ e see front matter ª 2011 Elsevier Ltd and IIR. All rights reserved. doi:10.1016/j.ijrefrig.2011.05.003

Upload: giuseppe

Post on 09-Nov-2015

222 views

Category:

Documents


3 download

DESCRIPTION

ANN 01_5-9-2

TRANSCRIPT

  • Available online 13 May 2011

    hot storage tanks, an auxiliary heater as well as two cold storage

    rmt en

    Mots cles : accumulation thermique ; syste`me a` absorption ; reseau neuronal

    Solar air conditioning is an emerging market with a huge

    growth potential. Peak cooling demand in summer is associ-

    ated with high availability of solar radiation, which offers an

    excellent opportunity to exploit solar energy with heat-driven

    cial technologies are available on a limited basis. However,

    a strong attention is put on research on other applications

    including photovoltaic-operated refrigeration cycles and solar

    mechanical refrigeration (Balaras et al., 2006). Recently

    research has been devoted to improving the main

    * Corresponding author. Tel.: 34 950 015914; fax: 34 950 015477.

    ava i lab le at www.sc iencedi rec t .com

    : w

    i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 4 ( 2 0 1 1 ) 1 4 4 6e1 4 5 4E-mail address: [email protected] (F.J. Batlles).1. Introduction cooling machines. For low power cooling systems, commer-de bacs daccumulation employant des reseaux neuronauxartificielsKeywords:

    Thermal storage

    Absorption system

    Neural network

    Etude sur la perfodair fonctionnan0140-7007/$ e see front matter 2011 Elsevdoi:10.1016/j.ijrefrig.2011.05.003out the study about the necessity of the integration of hot water storage tanks to solar

    system. Subsequently, the unique Artificial Neural Network (ANN) model with the lowest

    number of input variables has been proposed with the main purpose to predict the coef-

    ficient of performance and the cooling capacity of the absorption chiller. The configuration

    5-9-2 (5 inputs, 9 hidden and 2 output neurons) was found to be the optimal topology. The

    results demonstrate proper ANNs predictions with a Root Mean Square Error (RMSE) of less

    than 0.70% and practically null deviation, which can be considered very satisfactory.

    2011 Elsevier Ltd and IIR. All rights reserved.

    ance dun syste`me de conditionnementpartie grace a` lenergie solaire et muniAccepted 2 May 2011 tanks. The hot storage tank circuit was further investigated. In first step, we have carriedReceived in revised form

    29 March 2011

    system consists main

    a cooling tower, twoPerformance study of solar-assisted air-conditioning systemprovided with storage tanks using artificial neural networks

    S. Rosiek, F.J. Batlles*

    Dpto. Fsica Aplicada, Universidad de Almera, Ctra. Sacramento s/n, La Canada de San Urbano, 04120 Almera, Spain

    a r t i c l e i n f o

    Article history:

    Received 3 December 2009

    a b s t r a c t

    This study presents the performance of solar-assisted air-conditioning system provided

    with two storage tanks installed in the Solar Energy Research Center (CIESOL) building. The

    ly of solar collectors array, a hot-water driven absorption chiller,www. i ifi i r .org

    journa l homepageier Ltd and IIR. All rightsww.e lsev ier . com/ loca te / i j re f r igreserved.

  • T2 second storage tanks average temperature [C]

    i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 4 ( 2 0 1 1 ) 1 4 4 6e1 4 5 4 1447components of the solar absorption cooling system, such as

    the solar collector, the absorption chiller, and hot water

    storage tank.

    The main purpose of storage in a solar-assisted air-condi-

    tioning system is to overcome mismatches between solar

    gains and cooling loads. The most common application is the

    integration of a hot water buffer tank in the heating cycle of

    the thermally driven cooling equipment. The excess solar heat

    can be stored in the heat storage unit and is available if the

    solar heat is not sufficient and it serves as a buffer reservoir to

    have nearly constant heat input. Kreider and Kreith (1981)

    have reported that the system can be operated more

    efficiently by using two hot storage units as a heat reservoir,

    set at two different temperature ranges (Li and Sumathy,

    2000, 2001; Henning, 2004).

    Artificial Neural Networks have been increasingly used in

    recent years to predict or to improve nonlinear system

    performance in HVAC&R (S encan et al., 2006; Wong et al.,

    2010). The ability to learn is one of the outstanding charac-

    teristics on an ANN. ANNs can model multiple parameters

    simultaneously for nonlinear systems and are now widely

    used for predictive control, such as solar radiation (Lopez

    et al., 2001, 2005; Bosch et al., 2008), energy use prediction,

    energy optimization (Chow et al., 2002), data trending, and

    Nomenclature

    Cp specific heat capacity of water [4.18 Kj kg1 K1]

    I incident radiation intensity [Wm2]Tamb ambient air temperature [C]Tout leaving flat-plate collectors temperature [C]mc collectors mass flow rate [m

    3 h1]mg generators mass flow rate [m

    3 h1]Teg entering generators temperature [C]Tlg leaving generators temperature [C]me evaporators mass flow rate [m

    3 h1]

    Tee entering evaporators temperature [C]Tle leaving evaporators temperature [C]mac absorbers and condensers mass flow rate

    [m3 h1]optimum start and stop (Wang, 2001; Chang, 2007).

    There had been several studies on the performance

    prediction of the absorption chiller system (Lazzarin et al.,

    1993; Martnez and Pinazo, 2002; Syed et al., 2005; Asdrubali

    and Grignaffini, 2005; Helm et al., 2009), however very little

    work has been conducted for the total solar cooling systems

    based on absorption chiller and provided with hot water

    storage tanks. Hence the main purpose of this study was to

    analyze the essential importance of the hot water tank inte-

    gration and to determine ANN model able to estimate the

    coefficient of performance and cooling capacity for usage in

    control system. The final goal of the solar-assisted air-condi-

    tioning system is to operate with stable and high values of the

    coefficient of performance, and to maximize the use of solar

    thermal energy. To meet this objective we need to know in

    which conditions our systems works more efficiently, at the

    same time covering building necessity. Until now the control

    system has been utilized only by determining the information

    about temperatures found in the collectors field, storage tanksand the load fraction in the building with any consideration

    about the systems optimal operation points. This is the

    fundamental difference between systems with and without

    application of ANN models. On the other hand we intend to

    estimate the cooling capacity in order to operate with cold

    water storage tanks. Using this kind of tanks we are able to

    store the excess cooling capacity of the chiller and use it when

    thecoolingproductiondoesnot cover thebuilding load (clouds

    alternation or mismatches between solar gains and building

    loads). This system could permit the absorption chiller to be

    operated even when there is no demand, increasing the use of

    solarenergy,preventing thesuddenstart/stop (on/off cycles) of

    the chiller due to low cooling demand and allowing to realize

    the earlier chiller start-up, so we need to predict values of the

    cooling capacity to choose the best way of system control. In

    this way we maximize the use of solar thermal energy, avoid-

    ing the CO2 emissions to the atmosphere. In this work we

    determined a methodology, which could be easily adapted to

    other systems with the main goal to improve the design of

    solar-assisted systems bearing in mind number of measure-

    ments and operation points as well the maximisation of the

    solar thermal energy. Our study allows also reducing the

    monitoring cost, since we are able to avoid a lot of redundant

    measurementspoints through the selectionof theANNs input

    S2 temperature in the upper part of the second

    storage tank [C]COP coefficient of performance

    Qcool cooling capacity [kW]

    Qev the evaporator load [kW]

    Qgen the heat delivered to generator [kW]Teac entering absorbers and condensers temperature

    [C]Tlac leaving absorbers and condensers temperature

    [C]T1 first storage tanks average temperature [C]parameters. In the first phase of the present work we present

    the behaviour of the system without hot water storage tanks.

    To achieve this goal the study about the solar system with no

    hot water storage tanks integration was presented. Secondly,

    the solar-driven hot water storage tanks applied to above

    mentionedsystemwere investigatedmoreprofoundly. Finally,

    we determine an artificial neural networkwith themain scope

    to predict the coefficient of performance and the cooling

    capacity of the absorption chiller fed by water provided only

    from hot water storage tanks.

    2. Description of solar-assisted air-conditioning system

    In this study, we use data registered in the solar-assisted air-

    conditioning system installed in the Ciesol building situated

    on the Campus of the University of Almeria. The system

    employs a flat-plate solar collectors array with a total surface

  • of 160 m2, the hot-water driven single-effect LiBr-H2O

    absorption chiller with a rated capacity of 70 kW (Yazaki), the

    cooling tower, two hot storage tanks with a capacity of 5000 l

    each, an auxiliary heater and two cold storage tanks. Analysis

    of the aforementioned system and its various operation

    modes has been recently presented by Rosiek and Batlles

    (2009). Fig. 1 presents the view of the CIESOL building with

    the flat-plate solar collectors array installed on the roof and

    the main system components.

    In this study, we analyze the behaviour of this system

    supplied with only hot water storage tanks to satisfy the cool-

    ing demand of the Ciesol building. Fig. 2 presents the general

    scheme of this system operating in solar cooling mode. In the

    present paper, we use measurements of global radiation,

    temperatures and mass flow rates of the absorption chiller

    and storage tanks acquired with a 1 min sampling period

    with the main goal to analyze the behaviour of the system

    and to predict coefficient of performance (COP) and cooling

    capacity Qcool of the absorption chiller.

    The coefficient of performance (COP) is obtained from the

    following equation:

    COP Qev=Q

    gen m

    e Cp Tee Tle=m

    g Cp

    Teg Tlg

    (1)

    where Qev is the evaporator load Q

    cool,Q

    gen is the heat deliv-

    ered to generator, me is the evaporators mass flow rate, Cp is

    the specific heat capacity of water, Tee is the entering

    evaporators temperature, Tle, is the leaving evaporators

    temperature, mg is the generators mass flow, Teg is the

    entering generators and Tlg is the leaving generators

    temperature.

    3. Integration of the hot water storage tanks

    In the following subsections we want to emphasize the

    essential importance of thehotwater tank integration to solar-

    assisted air-conditioning system. To achieve this goal the

    study about the solar-assisted air-conditioning systemwithno

    hot water storage tanks integration was presented. Secondly,

    the solar-driven hot water storage tanks applied to above

    mentioned systemwere taken into deeper investigation.

    3.1. Solar cooling mode with no storage system

    In this paragraph we will study deeper the solar-assisted air-

    conditioning system driven only by solar energy with the

    main aim to underline the necessity of thermal storage inte-

    gration. The air-conditioning is in operation during office

    hours from 9 a.m. to 9 p.m., fromMonday to Friday. The basic

    idea of this mode is to feed the absorption chiller with water

    provided from solar collectors. The chiller has the minimum

    and maximum generators inlet temperature of the order of

    i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 4 ( 2 0 1 1 ) 1 4 4 6e1 4 5 414482Fig. 1 e a) View of the CIESOL building with 160m flat-plate sola

    facade, c) the nave with the main system components, d) hot wr collectors array, b) general view of CIESOL building-north

    ater storage tanks.

  • we can use some auxiliary heater for example, but the

    main purpose of this project is to maximize use of solar

    thermal energy, avoiding the CO2 emissions to the

    atmosphere, therefore the integration of the hot water

    storage tanks is essential. To underline once more the huge

    importance of integration of hot water storage tanks we

    attempted to show that the solar-assisted air-conditioning

    system could function only 28% of its total operating time

    (from 9 a.m. to 9 p.m.) working with no thermal storage or

    auxiliary heater (cf. Fig. 4).

    3.2. Solar cooling mode with hot storage system

    With the main goal to provide energy at moments when the

    energy proceeding from solar collectors is insufficient, two

    storage tanks were used. The use of a hot water tank between

    i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 4 ( 2 0 1 1 ) 1 4 4 6e1 4 5 4 144970 C and 95 C, respectively and out of those values the chillerwill stop working. Once the cooling operation is selected and

    the leaving flat-plate collectors temperature is greater than

    the minimum start-up temperature of absorption machine

    (70 C), the chiller will function automatically and remain inoperation as long as there is a demand for chilled water.

    During the period when absorption chiller is driven only by

    solar energy valve V1 is opened while both valves V2, V3 are

    kept closed and the pump P1 circulates the hot water between

    the absorption chiller and solar collectors array, avoiding

    storage tanks (c.f. Fig. 2). The bypass valve V4 is cycled ON and

    OFF to control flow of heat medium through the generator in

    response to the chilled water temperature. When the chilled

    water temperature is satisfied, the V4 is closed, the cooling

    water pump P2 will stopped after 4 min and the pump P3

    will circulate the chilled water through the fan-coil units.

    The system will remain in this state as long as the chilled

    water increases due to increasing buildings load,

    mg

    Flat-plate

    collectors array

    Absorption

    chiller

    Hot water

    pump

    P1

    Chilled water

    pump P3

    x

    Teg

    x

    Tlg

    me

    x

    x

    Tee

    Tle

    Incident radiation

    meter

    x Tout

    Cooling

    tower

    Cooling water

    pump P2

    xTsacTeac

    x

    mac

    Cooling

    load

    STORAGE TANKS

    T1 T2

    V2 V4V1V3

    Fig. 2 e The general scheme of the solar-assisted

    air-conditioning system driven by solar energy.meanwhile the valve V1 modifies its opening level

    depending on the difference between the entering and

    leaving flat-plate collectors temperature. The pump P1

    circulates the hot water between the absorption chiller and

    solar collectors and no excess energy can be store due to

    lack of the thermal storage.

    To verify the necessity of thermal storage integration the

    analysis of the leaving flat-plate collectors temperature has

    been carried out since the absorption chiller is driven only by

    solar energy. Taking into account the minimum generators

    inlet temperature we focussed our study only on the leaving

    collectors temperature superior than 70 C. The experi-mental data was collected during the cooling period of 2007

    and 2008 with 1 min sampling period. Fig. 3 presents the

    leaving flat-plate collectors temperature versus the incident

    radiation and collectors mass flow rate. As can be seen in

    the Fig. 3 for incident radiation less than 400 [Wm2] andcollectors mass flow rate higher than 9 [m3 h1] we cannotreach the leaving flat-plate collectors temperature higher

    than 70 C (the minimum start-up temperature ofabsorption machine), thus the auxiliary heat source is

    needed to start the cooling process. To fulfil this conditionFig. 3 e Leaving flat-plate temperature against collectorsthe solar collectors field and the absorptionmachine has been

    reported to yield higher system efficiency and extends the

    daily cooling period. It also prevents cycling of the absorption

    machine due to variations in solar radiation intensity (Li and

    Sumathy, 2000; Syed et al., 2005).

    In this paragraph we want to point out the necessity of

    storage tank integration, so we will focus on those modes of

    systems operation that involve thermal storage. Energy can

    be charged, stored and discharged daily or weekly depending

    on the control strategy. In the morning and afternoon, when

    the solar water is not sufficient to cover the cooling necessity,

    the system is fed by water provided from hot storage

    tanks. The control system compares the temperature of the

    second storage tank to the minimum start-up temperature of

    absorption machine (70 C) and if this temperature is graterthe chiller is switched on. During this operation mode valves

    V1 is kept closedwhile valves V2, V3 are opened and the pump

    P1 circulates the hot water between the absorption chiller and

    storage tanks, avoiding solar collectors (cf. Fig. 2). In this way

    the thermal energy is removing from storage tanks. The

    process of thermal storage discharging will continue as long

    as the second storage tanks temperature decreases below

    the minimum generators inlet temperature. During the

    period when the buildings load is low (low evaporatorsmass flow rate and incident radiation intensity during the

    cooling period of 2007 and 2008.

  • experimental day (03/09/07) while the ambient temperature

    reached 35 C. As we can see till 9:22 a.m. the evaporatorsleaving temperature was stable and of the order of 30 C.The solar-assisted air conditioning system was switched on

    at 9:22 a.m. with the main goal to cover the cooling demand

    in Ciesol building. The control system compared the

    temperature of the second storage tank and the leaving

    collectors to the minimum start-up temperature of the

    absorption chiller. At that moment the solar hot water was

    still too low and the warm water form storage tank was

    used to driven the chiller. From 9:22 a.m. to 10 a.m. the

    entering generators temperature presents the same profile

    as the second storage tank temperature, indicating that the

    absorption machine was provided with storage tanks.

    During this period the chilled water gradually decreased and

    reached the temperature of the order of 8 C. At the sametime the leaving collectors temperature increased sharply

    i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 4 ( 2 0 1 1 ) 1 4 4 6e1 4 5 41450temperatures) the control system demands low entering

    generators temperature. In those moments only the amount

    of solar heat necessary to cover the cooling loads is being

    used and the excess solar energy is stored in two hot water

    storage tanks. In this operation mode the pump P1 circulates

    warmer water from solar collectors to the first tank and at

    the same time the second storage tank is discharged to

    supply the absorption chiller (c.f. Fig. 2). Another way of

    storage tanks charging occurs when the chilled water is no

    needed (weekends and lunch break 2O4 p.m.) and hot water

    from solar collectors is used to load the thermal storage,

    since the absorption chiller is switched off. In this mode the

    Fig. 4 e Operation time of the solar cooling mode with no

    storage tank versus the total operation time in cooling

    period of 2007 and 2008.pump P1 circulates the hot water between solar collectors

    and storage tanks.

    Fig. 5 presents the temperature in the upper part of the

    second storage tank, leaving collectors temperature,

    generators entering temperature, ambient temperature and

    evaporators leaving temperature against time for one

    Fig. 5 e Generators entering temperature, evaporators leaving t

    the upper part of the second storage tank and ambient temperadue to experimental modification of V1 opening level. Even

    that the collectors temperature reached the value of about

    75 C which would be enough to drive the chiller, the controlsystem indicated the storage tanks as a heat source because

    of the high value of chilled water temperature. After 10 a.m.

    we can observe that the leaving evaporators temperature

    remains constant almost rest of the time due to low

    buildings load. From 10 a.m. to 2 p.m. only amount of solar

    heat necessary to cover the cooling demand was used and

    the excess solar energy was stored in two hot water storage

    tanks. Between 2 and 4 p.m. (lunch break) cold water was no

    needed and solar water was used to load the hot water

    storage tanks and it can be seen from Fig. 5 that the second

    storage tank temperature increased from 75 C to about85 C. After 4 p.m. the chiller once again was switched on.Once more the control system compared the temperature of

    the second storage tank and the leaving collectors to the

    minimum start-up temperature of the absorption chiller.

    Taking into account that the buildings load was low and the

    control system demanded low entering generators

    temperature, the absorption chiller was supplied with hot

    water from second storage tank. At the same time the first

    storage tank was charged with the hot water from solar

    collectors. The system operated till 6:15 p.m.emperature, leaving collectors temperature, temperature in

    ture against time (03/09/07).

  • Fig. 6 presents COP, obtained by applying Eq. (1), against

    entering generators temperature, leaving evaporators

    temperature and time for one experimental day (16/08/07).

    During this operation day we had 100% of the buildings load

    and the stable entering generators temperature of about

    70 C. Although the buildings load was high (highevaporators temperatures) and the control system should

    demand high entering generators temperature we set up the

    optimum entering generators temperature of about 70 Cwith the main goal to maximize the use of storage system.

    The solar-assisted air conditioning system was switched on

    at 12:33 p.m. and operated till 5:00 p.m. and during this

    period the COP varied from 0.31 to 0.84 at the beginning and

    at the end of the cooling process, respectively. We can also

    observe that the behaviour of the COP is almost constant

    during all experiment. It is mainly due to the fact that the

    absorption chiller was fed by water provided from second

    storage tank while the warmer water from solar collectors

    was used to charge the first tank. This operation mode

    stabilizes the behaviour of the COP and prevents cycling of

    neurons are set in layers, and thus a network is formed. Inputs

    representing the variables that affect the output of the

    network are feeding forward to each of the neurons in the

    following layers with activation depending on their weighted

    sum. Finally, an output can calculated as a function of the

    weighted sum of the inputs and an additional factor, the

    biases. The ability to learn is one of the outstanding charac-

    teristics of an ANN. The weights of the inputs are adjusted to

    produce a predicted output within specified errors. An ANN

    system is characterized by its net topology, neuron activations

    transfer, and learning method (Wang, 2001).

    Theneuralnetworkselectedhere isamultilayer feedforward

    perceptron (MLP) with one hidden layer. A tan-sigmoid transfer

    function was used as the activation function for the hidden

    layer, and a linear transfer function was used for the output

    layer. The LevenbergeMarquardt (LM) algorithmwas applied as

    themethod for achieving fast optimization (Chow et al., 2002).

    4.2. ANN model development and results

    i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 4 ( 2 0 1 1 ) 1 4 4 6e1 4 5 4 14514.1. ANN model description

    An artificial neural network (ANN) is massive interconnected,

    parallel processing, dynamic system of interacting processing

    elements that are in some aspect similar to the human brain.

    The fundamental processing element is called the neuron,

    which is analogous to the neural cell in human brain. Thethe absorption chiller that we can observe between

    12:58O1:41 p.m. and 2:18O2:32 p.m. In those moments the

    absorption chiller was provided with solar collectors and the

    entering generators temperature varied due to variations in

    solar radiation intensity. On other occasions we can observe

    that the storage tanks have an essential importance of the

    solar-assisted air-conditioning system.

    4. ANN modelFig. 6 e Coefficient of performance of the absorption chiller agai

    temperature and time (16/08/07).To determinate the ANN solar-assisted air-conditioning

    systemsmodel provided from storage tanks, a set of 1250 data

    points with a 1 min sampling period was used. Table 1

    presents the input and output parameters used for training

    the ANN absorption systems model. The input parameters

    were monitored by data acquisition system from June 6,

    2007 to September 25, 2007 and from July 14, 2008 to

    September 15, 2008, and the outputs parameters were

    estimated through the equation (c.f. Eq. (1)). In order to carry

    out the network training, 1060 data patterns were used, and

    the remaining 190 patterns were used as the test data set.

    The training and testing data were normalized between

    0 and 1, using (c.f. Eq. (2)):

    xscaled x xmin=xmax xmin (2)where xmax and xmin are equal to themaximumandminimum

    recorded values for each variable x. In order to determine thenst entering generators temperature, leaving evaporators

  • the number of input variables was varied. The selection of

    the input variables was started with the configuration of 13

    variables, and we progressively decreased the number of

    variables by taking into account their importance. Finally,

    we chose entering and leaving generators temperature,

    entering and leaving evaporators temperature and the

    second storage tanks average temperature as the more

    favourable configuration of the network inputs.

    After the input and output model variables were fixed, the

    next step consists of determining the network architecture

    Table 1 e Input and output parameters used for ANNabsorption systems model.

    Input variables Range

    Entering generators temperature [C] 64.8e90.7Leaving generators temperature [C] 38.7e84.5Generators mass flow rate [m3 h1] 7.3e16.8Entering evaporators temperature [C] 8.65e30.55Leaving evaporators temperature [C] 3.62e36.14Evaporators mass flow rate [m3 h1] 9.18e10.4Entering absorbers and condensers temperature [C] 24.02e41.3Leaving absorbers and condensers temperature [C] 24.77e43.64Absorbers and condensers mass flow rate [m3 h1] 0.07e45.08Incident radiation intensity [Wm2] 31.14e818.4Leaving flat-plate collectors temperature [C] 19e85.42First storage tanks average temperature [C] 59.24e81.49Second storage tanks average temperature [C] 62.87e87.45Coefficient of performance 0e0.99

    Table 2 e RMSE errors of coefficient of performance andcooling capacity obtained during selection of inputsvariables.

    Input variables RMSE [%]

    COP Qcool

    Teg, Tlg, mg, Tee, Tle, m

    e, Teac, Tlac, m

    ac, I, Tout, T1, T2 1.61 1.63

    Teg, Tlg, mg, Tee, Tle, m

    e, Teac, Tlac, I, Tout, T1, T2 1.33 1.06

    Teg, Tlg, mg, Tee, Tle, m

    e, I, Tout, T1, T2 1.56 0.93

    Teg, Tlg, mg, Tee, Tle, I, Tout, T1, T2 2.26 2.33

    Teg, Tlg, mg, Tee, Tle, m

    e, I, Tout, T2 1.55 1.39

    Teg, Tlg, Tee, Tle, me, I, Tout, T1, T2 3.49 1.46

    Teg, Tlg, mg, Tee, Tle, m

    e, Teac, Tlac, m

    ac 2.36 1.23

    Teg, Tlg, mg, Tee, Tle, m

    e, I, Tout, T1 2.33 2.05

    Teg, Tlg, mg, Tee, Tle, m

    e, Teac, I, T2 1.78 1.67

    Teg, Tlg, Tee, Tle, I, Tout, T1, T2 3.43 2.34

    Teg, Tlg, mg, Tee, Tle, m

    e, Teac, Tlac 1.83 1.80

    i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 4 ( 2 0 1 1 ) 1 4 4 6e1 4 5 41452ANN systems model, the neural network toolbox under the

    Matlab environment was used.

    When a large number of variables are eligible to be

    included in a model, selecting optimal inputs becomes a crit-

    ical step prior to the model development itself, since compu-

    tational cost can be considerably reduced and because not all

    the variables considered are always available (Bosch et al.,

    2008). In this study, we attempt to predict the performance

    of the absorption chiller provided from storage tanks with

    the main aim of lowering the initial input parameters.

    To verify the influence of the above mentioned input

    parameters, we considered the RootMean Square Error (RMSE)

    as a percentage of the mean measured values. In Table 2, one

    can see the statistical results of RMSE errors of coefficient of

    performance and cooling capacity obtained during selection

    of inputs variables for the ANN solar-assisted air-

    conditioning systems model.

    Themain goal of this work is to determine the unique ANN

    model with the minimal number of input patterns able to

    estimate the two output variables. Through the analysis of the

    results presented in the Table 2, we select themore favourable

    Cooling capacity [Kw] 0.46e79.34configuration of ANN model input variables. To carry out the

    network training, 1060 data patterns were used, where

    Fig. 7 e RMSE evolution vs. the increase of hidden units.Teg, Tlg, Tee, Tle, I, Tout, T1, T2 3.43 2.17

    Teg, Tlg, mg, Tee, Tle, m

    e, I, Tout 2.52 1.96

    Teg, Tlg, mg, Tee, Tle, m

    e, I, T2 2.18 1.71

    Tlg, mg, Tee, Tle, m

    e, I, Tout, T2 12.65 2.58

    Teg, Tlg, mg, Tee, Tle, m

    e, T2 2.46 1.80

    Teg, Tlg, mg, Tee, Tle, m

    e, Teac 1.75 1.58

    Teg, Tlg, mg, Tee, Tle, m

    e, Tlac 2.91 2.12

    Teg, Tlg, mg, Tee, Tle, m

    e 4.40 3.77

    Teg, Tle, I, Tout, T1, T2 25.52 24.6

    Teg, Tlg, Tee, Tle, I, T2 4.68 3.21

    Teg, Tlg, mg, Tee, Tle 4.22 3.41

    Teg, Tlg, Tee, Tle, me 3.48 1.87

    Teg, Tlg, Tee, Tle, T2 3.98 2.59

    Teg, Tlg, Tee, Tle 4.63 3.24(Bosch et al., 2008). Several MLPs networks with different

    numbers of hidden neurons Nh were trained. In order to

    Fig. 8 e MBE evolution vs. the increase of hidden units.

  • assess the accuracy of the neuralmodelswe analyze the results

    inmeaning of theRMSE andMeanBias Error (MBE) expressed as

    values are equal to 1 while the intercept values are very small.

    We can observe that the training values resulting in a good

    match to the experimental values.

    Fig. 9 e ANN architecture used for the absorption chiller

    system provided with two storage tanks.

    Fig. 11 e Comparison of actual and ANN-predicted values

    i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 4 ( 2 0 1 1 ) 1 4 4 6e1 4 5 4 1453apercentageof themeanmeasured. Figs.7and8 illustrateRMSE

    and MBE evaluation versus the increasing number of hidden

    neurons, respectively. As can be seen for numbers of hidden

    neurons higher than 9, the RMSE became almost constant,

    and the maximum and minimum deviations were found 0.023

    and -0.034, respectively. Finally, the architecture 5-9-2 (5

    inputs, 9 hidden and 2 output neurons) appears to be the most

    optimal topology.

    Fig. 9 presents the configuration of the two-layer back

    propagation network selected in this work. The input layer

    includes Teg, Tlg, Tee, Tle, and T2. The hidden layer has nine

    nodes, and the output layer includes COP andQcool.

    Once the training process was finished, we proceeded to

    compare the predicted values from ANN model with actual

    values. We used a set of 190 patterns as the test data. The

    accuracy of the ANN model was evaluated on the basis of theFig. 10 e Comparison of actual and ANN-predicted values

    of COP for the absorption chiller provided with two storage

    tanks.regression analysis of estimated versus measured values, in

    terms of the slope e a, intercept e b of the linear fit, the

    determination coefficient - R, RMSE and MBE. Figs. 10 and 11

    present the comparison between the actual and predicted

    values of COP and Qcool, for absorption system, respectively.

    The actual COP and Qcool were calculated as explained in

    paragraph 2. The aforementioned above figures show that the

    majority of the experimental points are located over the

    perfect adjust line 1:1, illustrating minimal dispersion.

    The RMSE error caused by the network in every case is less

    than 0.70%, and the deviation is practically null. It is noted

    that R values are equal and reach around 0.99, and all the slope

    of cooling capacity for the absorption chiller provided with

    two storage tanks.5. Conclusions

    The main objective of this study was to emphasize the great

    importance of hot water storage tanks integration in a solar-

    assisted air-conditioning system. The storage tanks system is

    very useful not only in the clods alternation when the solar

    hot water temperature is too low to cover the absorption

    machine demand but especially in the starter moments in the

    morning and whenever the incident radiation is minor than

    400 [Wm2]. It has been observed the solar-assisted air-conditioning system could function only 28% of its total

    operating time (from9 a.m. to 9 p.m.) workingwith no thermal

    storage. In this work, real data of an operating solar-assisted

    air-conditioning system provided from storage tanks was

    used to derive an ANN systems model to predict the coeffi-

    cient of performance and cooling capacity of the absorption

    chiller. The main aim of the present study was to determine

    the unique ANN model with a minimal number of input

    variables. Finally, the total of five variables were used as the

  • more favourable configuration of the network inputs -

    entering and leaving generators temperature, entering and

    leaving evaporators temperature and the second storage

    tanks temperature. Results demonstrate accurate predictions

    from the ANN model, yielding an RMSE less than 0.70% and

    practically null deviation, which can be considered very

    satisfactory. Results obtained through this prediction are very

    useful for control and monitoring strategies. The selection of

    the ANNs inputs permits reduction of the monitoring costs,

    since we are able to avoid a lot of redundant measurements

    points. In this work we determined a methodology, which

    could be easily adapted to other systemswith themain goal to

    improve the design of solar-assisted systems making them

    more attractive to potential users. Considering obtained

    acceptable results, we point out that future study in this field

    should focus on the use of the artificial neural networks to

    (MEC). The authors would like to thank to all companies and

    Chang, Y. Ch., 2007. Sequencing of chillers by estimating chillerpower consumption using artificial neural networks. Build.Envi. 42, 180e188.

    Chow, T.T., Zhang, G.Q., Lin, Z., Song, C.L., 2002. Globaloptimization of absorption chiller system by genetic algorithmand neural network. Energy Build. 34, 103e109.

    Helm, M., Keil, C., Hiebler, S., Mehling, H., Schweigler, C., 2009.Solar heating and cooling system with absorption chiller andlow heat storage: energetic performance and operationalexperience. Int. J. Refrigeration 32, 596e606.

    Henning, H.M., 2004. Solar-assisted Air-Conditioning in Buildings.A Handbook for Planners. Springer Wieb, New York.

    Kreider, J.F., Kreith, F., 1981. Solar system for space cooling. In:Solar Energy Handbook. McGraw-Hill, New York.

    Lopez, G., Rubio, M.A., Martinez, M., Batlles, F.J., 2001. Estimationof hourly global photosynthetically active radiation usingartificial neural network models. Agric. For. Meteorol. 107,279e291.

    i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 4 ( 2 0 1 1 ) 1 4 4 6e1 4 5 41454InfanteFerreira,C.A., 2006. Solar cooling.AnoverviewofEuropeanapplications and design guidelines. ASHRAE J., 48. June 2006.

    Bosch, J.L., Lopez, G., Batlles, F.J., 2008. Daily solar irradiationestimation over a mountainous area using artificial neuralnetworks. Renew. Energy 33, 1622e1628.institutions included in PSE-ARFRISOL project.

    r e f e r e n c e s

    Asdrubali, F., Grignaffini, S., 2005. Experimental evaluation of theperformances of a H2O-LiBr absorption refrigerator underdifferent service conditions. Int. J. Refrigeration 28, 489e497.

    Balaras, C.A., Henning, H.M.,Wiemken, E., Grossman, G., Podesser, E.,predict the global systems efficiency and energy usewhen the

    system works at all operation modes, considering the energy

    provided from all heat sources.

    Acknowledgements

    This research has been carried out with the help of the project

    PSE-ARFRISOL (The Singular Strategic Project called Biocli-

    matic Architecture and Solar Cooling), PS-120000-2005-1

    financed by the Spanish Ministry of Education and ScienceLopez, G., Batlles, F.J., Tovar-Pescador, J., 2005. Selection of inputparameters to model direct solar irradiance by using artificialneural networks. Energy 30, 1675e1684.

    Lazzarin, R.M., Romagnoni, P., Casasola, L., 1993. Two years ofoperation of a large solar cooling plant. Int. J. Refrigeration 16,185e190.

    Li, Z.F., Sumathy, K., 2000. Technology development in the solarabsorption air-conditioning systems. Renew. Sust. Energ. Rev.4, 267e293.

    Li, Z.F., Sumathy, K., 2001. Experimental studies on a solarpowered air conditioning system with partitioned hot waterstorage tank. Sol Energy 71, 285e297.

    Martnez, P.J., Pinazo, J.M., 2002. A method for design analysis ofabsorption machines. Int. J. Refrigeration 25, 634e639.

    Rosiek, S., Batlles, F.J., 2009. Integration of the solar thermalenergy in the construction: analysis of the solar-assisted air-conditioning system installed in Ciesol building. Renew.Energy 34, 1423e1431.

    S encan, A., Yakut, K.A., Kalogirou, S.A., 2006. Thermodynamicanalysis of absorption systems using artificial neural network.Renew Energy 31, 29e43.

    Syed, A., Izquierdo, M., Rodrguez, P., Maidment, G., Missenden, J.,Lecuona, A., Tozer, R., 2005. A novel experimentalinvestigation of a solar cooling system in Madrid. Int. J.Refrigeration 28, 859e871.

    Wang, S.K., 2001. Handbook of Air Conditioning andRefrigeration, second ed. McGraw-Hill, New York.

    Wong, S.L., Wan, K.K.W., Lam, T.N.T., 2010. Artificial neuralnetworks for energy analysis of office buildings withdaylighting. Appl. Energ 87, 551e557.

    Performance study of solar-assisted air-conditioning system provided with storage tanks using artificial neural networks1 Introduction2 Description of solar-assisted air-conditioning system3 Integration of the hot water storage tanks3.1 Solar cooling mode with no storage system3.2 Solar cooling mode with hot storage system

    4 ANN model4.1 ANN model description4.2 ANN model development and results

    5 Conclusions Acknowledgements References