Research ArticleAdaptive Chebyshev Neural Network Control for VentilatorModel under the Complex Mine Environment

Ranhui Liu , Xinyan Hu , Chengyuan Zhang, and Chuanxi Liu

Intelligent Equipment Academy, Shandong University of Science and Technology, Tai’an 271019, China

Correspondence should be addressed to Ranhui Liu; liuranhui@sdust.edu.cn

Received 5 July 2020; Accepted 21 July 2020; Published 27 August 2020

Guest Editor: Shubo Wang

Copyright © 2020 Ranhui Liu et al. %is is an open access article distributed under the Creative Commons Attribution License,which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Ventilator is important equipment for mines as it safeguards the lives under the shaft and ensures other equipment’s properfunctioning by providing fresh air. %erefore, how to effectively control the ventilator system becomes more significant. In orderto acquire the commonly used model and control strategy for ventilator systems, a new universal ventilator model is establishedbased on the blast capacity differential pressure in the ventilating duct and the ventilator motor model. %en, an adaptiveChebyshev neural network (ACNN) controller is proposed to effectively control the ventilator system where the unknown loadtorque and the unknown disturbance caused by the complex environment under the shaft are approximated by the Chebyshevneural network (CNN). Afterwards, an appropriate Lyapunov function candidate is designed to guarantee the stability of theproposed controller and the closed-loop ventilator system. Finally, the ACNN controller has been demonstrated to be effective interms of validity and precision for the new proposed ventilator model through the simulations.

1. Introduction

%e ventilator motor is very important for the mine since it isthe core equipment supporting the fresh air under the shaft[1–7]. Controlling the ventilator motor is always the mostimportant issue in the process of mining. In order to preciselycontrol the ventilator for saving the energy, the ventilatormodel must be investigated firstly. It has several kinds ofmodels (ARMAmodel, computational fluid dynamics model,etc.) for ventilator systems so far, but all are suiting specialconditions in special systems. %en, investigating the uni-versal ventilator system model for mines is of significance.Zhang et al. [2] investigated a train tunnel model based on bigdata. %e ventilation system was adjusted and regulated byiterative learning control algorithm and the results demon-strated the proposed control method was superior to activecontrol. Roubłk et al. [6] investigated amechanical ventilationfor lung injury aimed at preservation of spontaneousbreathing when using high-frequency oscillatory (HFO)ventilators, who designed a linear quadratic Gaussian statefeedback controller to compensate for the changes in meanairway pressure (MAP) in the ventilator circuit caused by

spontaneous breathing. Different from Zhang et al. [2] andRoubłk et al. [6], Yan et al. [7] investigated the high-altitudeventilator motor model for stratospheric airship airbags.%eyextended the traditional minimum stator current operation tominimize conduction below the rated speed concept, wherean extended minimum stator current operation (EMSCO)was designed at any speed to formalize as constrained op-timization problem. %en, they proposed optimal duty cyclemodel predictive current control (ODC-MPCC) to reduce thecurrent ripple and computation burden. From these refer-ences, one can find that most control strategies were based onoptimal control. But the mine environment even under theshaft is so intricate that many elements can disturb the op-timal control strategy. %erefore, the optimal controller willnot acquire the desired results under the complex disturbancein the mine. It needs a new control strategy for the ventilatormotor suiting the complex environment of the mine, espe-cially under the shaft.

Adaptive control is an intelligent control strategy whichcan solve most nonlinear complex issues [8–18]. In thispaper, we will adopt adaptive control for the ventilatormotor to restrain the strong nonlinear influence caused by

HindawiComplexityVolume 2020, Article ID 9861642, 10 pageshttps://doi.org/10.1155/2020/9861642

mailto:liuranhui@sdust.edu.cnhttps://orcid.org/0000-0002-8677-3973https://orcid.org/0000-0002-5017-5277https://creativecommons.org/licenses/by/4.0/https://doi.org/10.1155/2020/9861642

the complicated, noisy, and dusty environment of the mine.%ere are a lot of researchers applying the adaptive control tosolve the nonlinear issues. Chen et al. [19] proposed a fixed-time adaptive control for buck DC/DC converters. Anadaptive update law was adopted to estimate all the un-known parameters of the buck converters, and the fixed-time sliding mode surface (SMS) and corresponding controlwere developed for the buck converters with known pa-rameters such that the output error would converge to theequilibrium point within a fixed time. Gao et al. [14] in-vestigated the discrete Hammerstein system with Preisachhysteresis nonlinearity and unknown order linear dynamics.%ey proposed a lower triangular matrix to identify thePreisach density function and adopted Hankel matrix es-timating the linear dynamics order. %en, a compositecontrol consisting of discrete inverse model-based controller(DIMBC) and discrete adaptive sliding mode controller(DASMC) was proposed to deal with the Hammersteinsystem which can reduce the reaching time of DASMC andimprove the robustness of DIMBC. Sun et al. [16] designed afeedback adaptive controller for controlling the backlashservo system. Firstly, an extended state observer (ESO) wasadopted to estimate the system states and the vibrationtorque. %en, considering the feedback and feedforwardsignals, an adaptive robust compensation controller wasdesigned for this backlash servo system and the experimentresults have demonstrated the effectiveness for the systems.Different from former researchers, Wang et al. [17] appliedthe adaptive method to estimate the unknown parametersrather than using it as controller. %en, considering theproblem for the sluggish convergence caused by the onlinelearning, this study proposed a new adaptive law to achieveoptimal parameter estimation. In order to achieve thispurpose, an auxiliary filter was designed to compel theadaptive law with a time-varying gain under a cost function.Finally, an adaptive nonsingular terminal sliding modecontrol (ANTSMC) was proposed for a considered servosystem to obtain tracking control and parameter estimationsimultaneously. Na et al. [20] considered an energy maxi-mization problem for wave energy converters (WECs)subject to nonlinearities and constraints. %e adaptive ap-proach was used to solve the Hamilton–Jacobi–Bellmanequation and a critic neural network (NN) to approximatethe time-dependent optimal cost value. %ey combined theadaptive control and optimal control to deal with the waveenergy converters (WECs). From these studies, it shows thatthe adaptive control can perfectly be adopted for the non-linear systems with adaptive update law; even so, someunknown nonlinear sections or disturbances or models willalso be estimated by other methods, such as fuzzy logic [21]and neural network [22–25].

%e Chebyshev neural network (CNN) is a functional-link neural network that is a single-layer neural structurewith less computation to estimate the nonlinear function[26–30]. It has been utilized to estimate most kinds ofuncertain linear or nonlinear sections. Zou et al. [26] usedthe Chebyshev neural network for approximating thespacecraft attitude motion in uncertain spacecraft systems.%ey proposed Chebyshev neural network controller-I and

adaptive NN controller-II for global representation withoutsingularities of the systems. Different from Zou et al. [26],Gao and Liu [27] investigated a backlash-like hysteresisnonlinear system using the multiscale Chebyshev neuralnetwork (MSCNN) method to identify the unknown hys-teresis nonlinearity. Firstly, the unknown hysteresis back-lash-like nonlinearity would be approximated by the newmultiscale Chebyshev neural network. %en, the trackingerror was transformed into scalar error by Laplace trans-formation to simplify the computation. Finally, an adaptivecontrol was designed for controlling the backlash-likehysteresis nonlinear system, and the simulations verified theeffectiveness of the multiscale Chebyshev neural networkidentification and the adaptive controller. Sun et al. [28]focused on the multimotor servomechanism with unmod-eled dynamics. An extended state observer based on high-order sliding mode (HOSM) differentiator was applied forunmeasured velocity, and the Chebyshev neural networkwas adopted to deal with the friction and disturbances. %eexperiments proved that the proposed approach was useful.To sum up, the Chebyshev neural network was utilized fordifferent nonlinear systems and obtained good results.%erefore, we will utilize the CNN to deal with the unknownnonlinear complex uncertain sections in the ventilatorsystem in this paper.

Firstly, a new ventilator model will be established basedon the blast capacity differential pressure of the ventilatingduct and the ventilator motor, that is, a nonlinear dynamicmodel where some load torques and some complex dis-turbances are unknown. %en, a Chebyshev neural networkwill be designed to estimate the unknown nonlinear sectionand an adaptive CNN controller will be proposed for theclosed-loop ventilator system. Finally, the appropriateLyapunov function guarantees the convergence of theclosed-loop ventilator system and the simulations also affirmthat the proposed approach is precise and effective.

%is paper is organized as follows. Section 2 gives theproblem formulations, which introduces the proposed ven-tilator model based on the blast capacity differential pressureand the ventilator motor and also introduces the structure ofthe Chebyshev neural network. Section 3 presents the con-troller design, that is, the process to the design control. %estability of the ventilator system by Lyapunov function is alsogiven in this section. %en, simulations are given in Section 4,and Section 5 concludes the paper.

2. Problem Formulations

2.1. %e Ventilator Model. %e most important index of theventilator model is the flux of the ventilating duct. Con-sidering the complicated and dusty environment of theventilating duct in the mine, the approach of the blast ca-pacity differential pressure is selected to solve the fresh airvelocity measurement hole blocking in order to realize thereal-time accuracy measurement and monitor the windpressure. %e schematic diagram structure of the blast ca-pacity differential pressure method is illustrated in Figure 1.From Figure 1, the energy equations of sections I and II canbe deduced as follows:

2 Complexity

ps1 + ρ1Z1g +ρ1v

21

2� ps2 + ρ2Z2g +

ρ2v22

2+ hr,

ps1 + ρ1Z1g +ρ1v

21

2� ps2 + ρ2Z2g +

ρ2v22

2+ ξ

ρv222

,

⎧⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎩

(1)

where ps1, ps2, ρ1, ρ2, Z1, Z2, v1, v2 represent the static pres-sure, air density, elevation, and air speed of sections I and II,respectively, hr is the local resistance of section II, and ξ isthe coefficient of local resistance of section II.

Since sections I and II have close distance and sameelevation, ρ � ρ1 � ρ2 � Z1 � Z2 holds. Considering theequation (1), we have

ps1 − ps2 �ρv222

−ρv212

. (2)

Since sections I and II have close distance, the air volumeof ventilating duct Q can be deduced as

Q � Q1 � Q2, (3)

where Q1, Q2 represent the air volume of sections I and II,respectively.

Substituting v1 � Q/s1, v2 � Q/s2 into equation (2), weobtain

Q � s1s2

����������

2 ps1 − ps2( ρ s21 − s

22

,

s1 �πd214

,

s2 �π d22 − d

21

4,

(4)

where s1, s2 represent the areas with the different definitionin (4), respectively, and d1, d2 mean outer diameter andinsider diameter of the fairing, respectively. From equation(4), the ventilator air volume flow can be calculated whereρ, s1, s2, ps1, ps2 is measured by the different sensors. Basedon reference [7], it is obvious that the effective control forventilator motor will save energy and control the air volumeflow Q. %erefore, an effective control strategy for ventilatormotor will effectively control the ventilator air volume flow.%e ventilator motor model in D-Q coordinates is repre-sented as follows:

_d

€q

⎡⎢⎢⎢⎢⎢⎣⎤⎥⎥⎥⎥⎥⎦ �

−R

L

K

L

KtJi

1Ji

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

iq

_q

⎡⎢⎢⎢⎣ ⎤⎥⎥⎥⎦ +

uq

L

−Tn + Tl

L

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦, (5)

where R, L represent the stator resistance and inductance,respectively. K � nψ, n is the number of the pole pairs, ψ isthe rotor flux linkage, Kt, Ji, uq are the torque constant,inertia, and q-axis stator voltages, respectively. Tn is thetorque caused by some nonlinear portion such as friction,

disturbance, and so on. Tl � f(Q) is the load torque which isthe unknown function about ventilator air volume flow inthis paper.

Define x � [x1, x2]T � [q, _q]T; based on reference [31],

ventilator motor model (5) can be rewritten as

_x1

_x2

⎡⎢⎢⎢⎣ ⎤⎥⎥⎥⎦ �

0 1

0KKt

JiR

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

x1

x2

⎡⎢⎢⎢⎣ ⎤⎥⎥⎥⎦ +

0

Ktuq − TnR − Rf(Q)

JiR

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦,

y � x1.

(6)

2.2. ChebyshevNeuralNetwork. In this section, a CNN willbe designed to estimate the unknown sections in equa-tion (6). CNN is one of the functional-link neural net-works whose structure is illustrated in Figure 2. FromFigure 2, CNN is a single-layer neural network whichrelies on the Chebyshev polynomials. %erefore, it haslower computation and is widely used for functionestimation.

%e CNN can be defined as follows:

F(x) � W∗Ψ(x) + ζ, (7)

where W∗ is the weight matrix of the CNN and ζ is thebounded approximation error for the CNN.

Ψ(x) � 1, d1 x1( , d2 x1( , . . . , dm x1( , d1 x2( , d2

· x2( , . . . , dm x2( , . . . , d1 xn( , d2 xn( , . . . , dm xn( T,

(8)

where [x1, x2, . . . , xn]T is the n-dimensional input vector, m

represents the order of the Chebyshev polynomial, and theChebyshev polynomial dk(xl) is defined as the two-termrecursive formula as follows:

dk xl( � 2xldk−1 xl( − dk−2 xl( , (9)

where the original value of d(x) is selected asd1(xl) � xl, d0(xl) � 1.

%e CNN update law is defined as follows:

W.

� −ϱ32

e2Ψ(x), (10)

where ϱ3 is the designed positive constant, e2 is defined in(16), and W is the estimation of the weight W∗.

3. Controller Design

Considering the structure of the ventilator models (1) and(6), we should design a recursive control strategy to solve theproblem. %e recursive controller method is designed asfollows.

Step 1: in order to design the U-model CNN adaptivecontroller, the output error is defined as follows:

e � y − yd, (11)

Complexity 3

where yd is a known signal which has (m + 1)th de-rivative such as a sinusoidal signal.Considering equation (6), the derivative of e can bededuced as

_e � _y − _yd� _x1 − _yd� x2 − _yd.

(12)

Assume Tn and f(Q) are unknown, where TnR andRf(Q) are estimated by the CNN. %en, the 2th de-rivative of e is obtained as

€e � _x2 − €yd

�KKt

JiRx2 +

Ktuq − TnR − Rf(Q)

JiR− €yd

�KKt

JiRx2 +

Kt

JiRuq − W

∗Ψ(x) − €yd.

(13)

Define the following Lyapunov function candidate as

V1 �12

e2. (14)

%en, the derivate of V1 can be deduced as_V1 � e _e � e x2 − _yd( . (15)

In this step, x2 is treated as a virtual input; therefore, wedefine an error as e2 � x2 − a2, where a2 is a virtualcontrol law. Substitute x2 � e2 − a2 into (15), and wehave

_V1 � e e2 − a2 − _yd( . (16)

Define the virtual control law a2 as follows:

a2 � l1e − _yd + e2, (17)

where l1 is a positive feedback control gain, and we cango to the next step.

x1 Ф (xk1)W (xk1)

Activation function

Sign()

Update lawError

Output

Desired output

–

+

W (xk2)

W (xkn)

Ф (xk2)

Ф (xkn)

x2

xn

Cheb

yshe

vex

pans

ion

Figure 2: %e structure of the Chebyshev neural network.

Ventilation duct

Section II

Differentialpressure

transmitterStatic

pressuretransmitter

Section I

Pressuretransmitter

Figure 1: %e structure of the blast capacity differential pressure.

4 Complexity

Step 2: the actual controller u will be acquired in thisstep. Firstly, we define the derivative of e2 as

_e2 � _x2 − _a2. (18)

Based on (17), the derivation of a2 is deduced as

_a2 � l1 _e − €yd + _e2

� l1 e2 − a2 − _yd( − €yd + _e2,(19)

and then substituting (19) into (18), one can get

_e2 �12

KKt

JiRx2 +

Kt

JiRuq − W

∗Ψ(x) − l1e2 + l1a2 + l1 _yd + €yd .

(20)

Define the adaptive update errors as

a2 � a2 − a2,

x2 � x2 − x2,

W � W∗

− W,

(21)

where a2, x2 are the estimations of a2, x2.%e adaptive update law is defined as follows:

_x2 �ϱ12

KKt

JiRe2,

_a2 �ϱ22

l1e2,

(22)

where ϱ1, ϱ2 are designed positive constants.We select the true controller input as

uq �JiR

Kt−

KKt

JiRx2 +

WΨ(x) − l1a2 − l1 _yd − €yd . (23)

Select the Lyapunov function candidate as follows:

V �12e22 +

12ϱ1

x22 +

12ϱ2

a22 +

12ϱ3

WT W. (24)

%e derivative of V is deduced as

_V � e2 _e2 +1ϱ1

x2_x2 +

1ϱ2

a2_a2 +

12ϱ3

WT _W. (25)

Substituting (20) into (25), the derivative of V can bededuced as

_V �e2

2KKt

JiRx2 +

Kt

JiRuq − W

∗Ψ(x) − l1e2 + l1a2 + l1 _yd + €yd

+1ϱ1

x2_x2 +

1ϱ2

a2_a2 +

12ϱ3

WT _W.

(26)

Substituting the true controller (23) into (26), one canobtain

_V �e2

2KKt

JiRx2 −

KKt

JiRx2 − W

∗Ψ(x) − l1e2 + l1a2 + l1 _yd

+ €yd +WΨ(x) − l1a2 − l1 _yd − €yd +

1ϱ1

x2_x2

+1ϱ2

a2_a2 +

12ϱ3

WT _W

� −12l1e

22 +

e2

2KKt

JiRx2 +

WΨ(x) + l1a2 −1ϱ1

x2_x2

1ϱ2

a2_a2

−1ϱ3

WT _W.

(27)

Substituting (10) and (22) into (27), according toLyapunov theory, one can obtain that the closed loop isstable with the designed CNN adaptive controller.

4. Simulation

%e simulations is designed to verify the effectiveness of theventilator motor in the mine. In the simulations, the ven-tilator model parameters are listed in Table 1. %e param-eters of the ventilator motor model in D-Q coordinates arelisted in Table 2. %e reference input is selected asyd � sin(t). %e simulation results are shown inFigures 3–12. From Figures 3–7, it is clearly shown that theproposed control strategy is effective.

Figure 3 shows the control results with the proposedneural network adaptive controller. %e controlled outputcan quickly track the reference input with smaller andconvergent error which is also shown in Figure 4. FromFigure 4, it is very obviously illustrated that the error isconverged and the mean absolute error (MAE) is less than0.02. Figure 5 shows the controller input, and it shows thesame conclusions compared with Figures 3 and 4. FromFigure 5, the dynamic process can be clearly shown less than1 second in these simulations, and then the neural networkand the adaptive update law can estimate the unknownnonlinear section and parameters quickly. %e trackingresult is clearly shown in Figure 3. Figures 6 and 7 show thetrajectories for the estimation x1, x2 of x1 and x2. From thesetwo diagrams, it can be concluded that the proposed methodcan estimate the system state rapidly and the trackingcontroller is effective, which are in line with the results inFigures 3–5.

To magnify the amplitude of the input signal, the resultsare shown in Figures 8–12. From Figures 8 and 9, one canhave the same conclusions compared with Figures 3 and 4.But comparing Figure 10 with Figure 5, it is clearly illus-trated that Figure 10 has more drastic change. %en,comparing Figures 3–5 with Figures 8–10, it is illustratedthat magnifying the signal amplitude will influence thecontrolled input, but the proposed controller also retains thecontrol precision with strong control input. Figures 11 and12 also show the same conclusion.

Complexity 5

Remark 1. In this paper, since the controller is designedbased on the output tracking error, the tracking controlresults are excellent in Figures 3 and 8. But in Figures 7 and12, the system states x1, x2 have a chattering. %e reason isthat we did not consider the state control and we also did notutilize the state observer. %en, the tracking error can affectthe system states by the feedback structure that makes somechattering in the system states. We will further investigatethis phenomenon in the future work.

0 5 10 15 20 25 30Time (s)

–0.04

–0.02

0

0.02

0.04

0.06

0.08

0.1

The e

rror

of t

he o

utpu

t

Figure 4: %e errors of output with small input signal amplitude.

0 5 10 15 20 25 30Time (s)

–40

–30

–20

–10

0

10

20

30

40

The c

ontr

ol in

put (

V)

Figure 5: %e trajectory of the controller input.

Table 1: %e parameters of the ventilator.

Parameters Value Unitsps1 1.51 kPaps2 1.52 kPaρ1 1 kg/m3ρ2 1 kg/m3Z1 −50 mZ2 −50 mv1 10 m/sv2 10 m/shr 150 Ns1 0.2 m2s2 0.001 m2ξ 0.8

Table 2: %e parameters of the ventilator motor in D-Qcoordinates.

Parameters Value UnitsJs 0.01K 0.5Kt 10R 0.1 kΩTn 0.02 Nm

0 5 10 15 20 25 30–1

–0.8

–0.6

–0.4

–0.2

0

0.2

0.4

0.6

0.8

1

Out

put

Times (s)

Desired outputOutput

Figure 3: %e control results with small input signal amplitude.

0 5 10 15 20 25 30Time (s)

–1.5

–1

–0.5

0

0.5

1

1.5

The e

stim

atio

n of

x1

1

1

Figure 6: %e estimation of x1.

6 Complexity

–25

–20

–15

–10

–5

0

5

10

15

20

The e

stim

atio

n of

x2

0 5 10 15 20 25 30Time (s)

Figure 7: %e estimation of x2.

–1

–0.8

–0.6

–0.4

–0.2

0

0.2

0.4

0.6

0.8

1

Out

put

0 5 10 15 20 25 30Time (s)

Desired outputOutput

Figure 8: %e control results with large input signal amplitude.

–0.04

–0.02

0

0.02

0.04

0.06

0.08

0.1

The e

rror

of t

he o

utpu

t

0 5 10 15 20 25 30Time (s)

Figure 9: %e errors of output with large input signal amplitude.

Complexity 7

–50

–40

–30

–20

–10

0

10

20

30

40

The c

ontro

l inp

ut (V

)

0 5 10 15 20 25 30Time (s)

Figure 10: %e trajectory of the controller input.

–1.5

–1

–0.5

0

0.5

1

1.5

The e

stim

atio

n of

x1

0 5 10 15 20 25 30Time (s)

1

1

Figure 11: %e estimation of x1.

–15

–10

–5

0

5

10

15

The e

stim

atio

n of

x2

0 5 10 15 20 25 30Time (s)

Figure 12: %e estimation of x2.

8 Complexity

5. Conclusion

A new adaptive Chebyshev neural network (ACNN) controlwas presented to precisely control the ventilator system inthis paper. Considering that the optimal controller does notsuit the complex environment under the shaft in the minewhich is easily affected by the strong disturbance caused bythe hostile environment, the ACNN replaced the optimalcontrol compared with other traditional control strategies.Firstly, the new ventilator model was proposed whichconsisted of the blast capacity differential pressure equationin the ventilator duct and the ventilator motor equation.%en, the CNN estimated the unknown nonlinear sectionssuch as the unknown load torque and the unknown dis-turbance caused by the hostile environment under the shaft.Finally, the stability of the closed-loop ventilator system wasguaranteed by a designed appropriate Lyapunov functioncandidate and the simulation results verified the effective-ness and the precision of the established new ventilatormodel and the presented ACNN control strategy.

Data Availability

No data were used to support this study.

Conflicts of Interest

%e authors declare that they have no conflicts of interest.

Acknowledgments

%is study was supported by the Shandong Provincial KeyResearch and Development Project (2019GGX101005),Shandong Provincial Natural Science Foundation(ZR2017MF048), and National Natural Science Foundationof China (61803216).

References

[1] A. Das, P. P. Menon, J. G. Hardman, and D. G. Bates,“Optimization of mechanical ventilator settings for pulmo-nary disease states,” IEEE Transactions on Biomedical Engi-neering, vol. 60, no. 6, pp. 1599–1607, 2013.

[2] Y. Zhang, J. Su, and M. Chen, “Research on adaptive iterativelearning control of air pressure in railway tunnel with IOTsdata,” IEEE Access, vol. 8, pp. 5481–5487, 2020.

[3] P. Casaseca-de-la-Higuera, F. Simmross-Wattenberg,M. MartIn-Fernandez, and C. Alberola-Lopez, “A multi-channel model-based methodology for extubation readinessdecision of patients on weaning trials,” IEEE Transactions onBiomedical Engineering, vol. 56, no. 7, pp. 1849–1863, 2009.

[4] H.-T. Wu, S.-S. Hseu, M.-Y. Bien, Y. R. Kou, andI. Daubechies, “Evaluating physiological dynamics via syn-chrosqueezing: prediction of ventilator weaning,” IEEETransactions on Biomedical Engineering, vol. 61, no. 3,pp. 736–744, 2014.

[5] R. Amini, J. Herrmann, and D. W. Kaczka, “Intratidaloverdistention and derecruitment in the injured lung: asimulation study,” IEEE Transactions on Biomedical Engi-neering, vol. 64, no. 3, pp. 681–689, 2017.

[6] K. Roubłk, J. Ráfl, M. van Heerde, and D. G. Markhorst,“Design and control of a demand flow system assuring

spontaneous breathing of a patient connected to an HFOventilator,” IEEE Transactions on Biomedical Engineering,vol. 58, no. 11, pp. 3225–3233, 2011.

[7] L. Yan, M. Dou, Z. Hua, H. Zhang, and J. Yang, “Optimal dutycycle model predictive current control of high-altitude ven-tilator induction motor with extended minimum statorcurrent operation,” IEEE Transactions on Power Electronics,vol. 33, no. 8, pp. 7240–7251, 2018.

[8] Y. Zhang and G. Li, “Non-causal linear optimal control ofwave energy converters with enhanced robustness by slidingmode control,” IEEE Transactions on Sustainable Energy, p. 1,2019.

[9] Q. Chen, H. Shi, and M. Sun, “Echo state network-basedbackstepping adaptive iterative learning control for strict-feedback systems: an error-tracking approach,” IEEE Trans-actions on Cybernetics, vol. 50, no. 7, pp. 3009–3022, 2020.

[10] S. Wang and J. Na, “Parameter estimation and adaptivecontrol for servo mechanisms with friction compensation,”IEEE Transactions on Industrial Informatics, vol. 16, no. 11,pp. 6816–6825, 2020.

[11] Y. Zhang, T. Zeng, and G. Li, “Robust excitation force esti-mation and prediction for wave energy converterM4 based onadaptive sliding-mode observer,” IEEE Transactions on In-dustrial Informatics, vol. 16, no. 2, pp. 1163–1171, 2020.

[12] Q. Chen, S. Xie, M. Sun, and X. He, “Adaptive nonsingularfixed-time attitude stabilization of uncertain spacecraft,” IEEETransactions on Aerospace and Electronic Systems, vol. 54,no. 6, pp. 2937–2950, 2018.

[13] S. Wang, L. Tao, Q. Chen, J. Na, and X. Ren, “USDE-basedsliding mode control for servo mechanisms with unknownsystem dynamics,” IEEE/ASME Transactions onMechatronics,vol. 25, no. 2, pp. 1056–1066, 2020.

[14] X. Gao, X. Ren, C. Zhu, and C. Zhang, “Identification andcontrol for hammerstein systems with hysteresis non-line-arity,” IET Control %eory Applications, vol. 9, no. 13,pp. 1935–1947, 2015.

[15] Y. Zhang, R. Vepa, G. Li, and T. Zeng, “Mars powered descentphase guidance design based on fixed-time stabilizationtechnique,” IEEE Transactions on Aerospace and ElectronicSystems, vol. 55, no. 4, pp. 2001–2011, 2019.

[16] G. Sun, J. Zhao, and Q. Chen, “Observer-based compensationcontrol of servo systems with backlash,” Asian Journal ofControl, 2019.

[17] S. Wang, J. Na, and Y. Xing, “Adaptive optimal parameterestimation and control of servo mechanisms: theory andexperiments,” IEEE Transactions on Industrial Electronics,p. 1, 2020.

[18] G. Xuehui, Z. Wei, W. Shubo, W. Minlin, and R. Xuemei, “Aprescribed performance adaptive control for hysteresisHammerstein system,” Journal of Systems Science & Com-plexity, vol. 32, no. 4, pp. 1039–1052, 2019.

[19] Q. Chen, N. Qian, and Y.-R. Nan, “Fixed-time adaptivecontrol for buck converters,” Kongzhi Yu Juece/Control andDecision, vol. 35, no. 5, pp. 1183–1190, 2020.

[20] J. Na, B. Wang, G. Li, S. Zhan, and W. He, “Nonlinearconstrained optimal control of wave energy converters withadaptive dynamic programming,” IEEE Transactions on In-dustrial Electronics, vol. 66, no. 10, pp. 7904–7915, 2019.

[21] J. Na, Y. Huang, X. Wu, S.-F. Su, and G. Li, “Adaptive finite-time fuzzy control of nonlinear active suspension systemswith input delay,” IEEE Transactions on Cybernetics, vol. 50,no. 6, pp. 2639–2650, 2020.

[22] S. Wang, H. Yu, J. Yu, J. Na, and X. Ren, “Neural-network-based adaptive funnel control for servo mechanisms with

Complexity 9

unknown dead-zone,” IEEE Transactions on Cybernetics,vol. 50, no. 4, pp. 1383–1394, 2020.

[23] X. Gao, “Adaptive neural control for hysteresis motor drivingservo system with Bouc-Wen model,” Complexity, vol. 2018,Article ID 9765861, 9 pages, 2018.

[24] B. Sun, C. Yang, Y. Wang, W. Gui, I. Craig, and L. Olivier, “Acomprehensive hybrid first principles/machine learningmodeling framework for complex industrial processes,”Journal of Process Control, vol. 86, pp. 30–43, 2020.

[25] J. Na, A. S. Chen, Y. Huang et al., “Air-fuel ratio control ofspark ignition engines with unknown system dynamics es-timator: theory and experiments,” IEEE Transactions onControl Systems Technology, pp. 1–8, 2019.

[26] A. M. Zou, K. Dev Kumar, and Z. G. Hou, “Quaternion-basedadaptive output feedback attitude control of spacecraft usingChebyshev neural networks,” IEEE Transactions on NeuralNetworks, vol. 21, no. 9, pp. 1457–1471, 2010.

[27] X. Gao and R. Liu, “Multiscale Chebyshev neural networkidentification and adaptive control for backlash-like hysteresissystem,” Complexity, vol. 2018, Article ID 1872493, 9 pages,2018.

[28] G. Sun, X. Ren, and D. Li, “Neural active disturbance rejectionoutput control of multimotor servomechanism,” IEEETransactions on Control Systems Technology, vol. 23, no. 2,pp. 746–753, 2015.

[29] M.-F. Leung and J. Wang, “A collaborative neurodynamicapproach to multiobjective optimization,” IEEE Transactionson Neural Networks and Learning Systems, vol. 29, no. 11,pp. 5738–5748, 2018.

[30] R. Levie, F. Monti, X. Bresson, and M. M. Bronstein, “Cay-leyNets: graph convolutional neural networks with complexrational spectral filters,” IEEE Transactions on Signal Pro-cessing, vol. 67, no. 1, pp. 97–109, 2019.

[31] S. Wang, X. Ren, J. Na, and T. Zeng, “Extended-state-ob-server-based funnel control for nonlinear servomechanismswith prescribed tracking performance,” IEEE Transactions onAutomation Science and Engineering, vol. 14, no. 1, pp. 98–108, 2017.

10 Complexity