mathematics

Article
Adaptive Backstepping Terminal Sliding Mode Control of
Nonlinear System Using Fuzzy Neural Structure
Xiaoyu Gong 1 , Wen Fu 2 , Xingao Bian 2 and Juntao Fei 1,2, *

1 College of IoT Engineering, Jiangsu Key Laboratory of Power Transmission and Distribution Equipment
Technology, Hohai University, Changzhou 213022, China
2 College of Mechanical and Electrical Engineering, Hohai University, Changzhou 213022, China
* Correspondence: [email protected]; Tel.: +86-519-8519-2023

Abstract: An adaptive backstepping terminal sliding mode control (ABTSMC) method based on a
multiple−layer fuzzy neural network is proposed for a class of nonlinear systems with parameter
variations and external disturbances in this study. The proposed neural network is utilized to
estimate the nonlinear function to handle the unknown uncertainties of the system and reduce
the switching term gain. It has a strong learning ability and high approximation accuracy due to
the combination of a fuzzy neural network and recurrent neural network. The neural network
parameters can be adaptively adjusted to optimal values through the adaptive laws derived from the
Lyapunov theorem. To stabilize the control signal, the additional parameter adaptive law derived
by the adaptive projection algorithm is used to estimate the control coefficient. The terminal sliding
mode control (TSMC) is introduced on the basis of backstepping control, which can ensure that the
tracking error converges in finite time. The simulation example is carried out on the DC–DC buck
converter model to verify the effectiveness and superiority of the proposed control method. The
contrasting results show that the ABTSMC−DHLRNN possesses higher steady−state accuracy and
faster transient response.

Keywords: multiple−layer fuzzy neural network; recurrent neural network; adaptive projection
algorithm; adaptive backstepping terminal sliding mode control
Citation: Gong, X.; Fu, W.; Bian, X.;
Fei, J. Adaptive Backstepping MSC: 68T07; 93C40
Terminal Sliding Mode Control of
Nonlinear System Using Fuzzy
Neural Structure. Mathematics 2023,
11, 1094. https://doi.org/ 1. Introduction
10.3390/math11051094
It is well known that most systems existing in a variety of scenarios are nonlinear
Academic Editor: systems with a strict feedback form. The control purpose is to ensure that the state signals
Sundarapandian Vaidyanathan of the system follow the reference signals. As the loading conditions and system parameters
Received: 27 January 2023
are highly uncertain, it will lead to an undesirable impact on system regulation accuracy.
Revised: 14 February 2023
Therefore, it is necessary to design a stable and adaptive controller for a nonlinear system
Accepted: 20 February 2023 with the presence of external disturbances and parameter variations.
Published: 22 February 2023 It can be seen from early research that the PID controller has been widely concerned
due to its simple principle and application. In [1], a PID controller was designed to suppress
current harmonics for a permanent−magnet synchronous motor (PMSM). Irfan et al. [2]
designed double PID controllers per axis for a stable flight of Vertical Take−Off and Land-
Copyright: © 2023 by the authors. ing drones. In order to handle the rapid variation in unknown disturbances, a dynamic
Licensee MDPI, Basel, Switzerland. parameter tuning method for variable PID controllers was proposed in [3]. Moreover,
This article is an open access article Meza et al. [4] presented a fuzzy self−tuning method to select PID gains according to the
distributed under the terms and actual state of the model. However, the PID controller is a linear control strategy, which
conditions of the Creative Commons seldom ensures large−signal stability and robustness. Hence, more advanced control strate-
Attribution (CC BY) license (https://
gies are proposed to handle the instability problem. In [5], a proportional−integral−fuzzy
creativecommons.org/licenses/by/
controller based on an iterative feedback tuning algorithm was designed to control a class
4.0/).

Mathematics 2023, 11, 1094. https://doi.org/10.3390/math11051094 https://www.mdpi.com/journal/mathematics

Mathematics 2023, 11, 1094 2 of 21

of servo systems. In recent research, some nonlinear control methods have been widely
used, including sliding mode control (SMC) [6] and backstepping control [7].
The main feature of backstepping control is that it designs virtual control laws for each
subsystem through a step−by−step recursive process, which can reduce the complexity of
the control design. Cai et al. [8] proposed a robust adaptive control strategy based on the
backstepping technique for nonlinear systems with unknown parameters, and Lin et al. [9]
designed an adaptive backstepping speed control system for a PMSM. However, when sys-
tem uncertainty and disturbances are involved, it is difficult for conventional backstepping
control to achieve high performance.
Due to the design of the switching term, SMC is another attractive choice for applica-
tion in the nonlinear system, which has the property of robustness to disturbances. In [10],
a second−order SMC algorithm was presented to solve motion control problems of robot
manipulators. In [11], a super−twisting algorithm was introduced into SMC, which can
effectively weaken the chattering phenomenon of the control input. Here, the backstep-
ping sliding mode control (BSMC) was proposed, which combines both characteristics of
backstepping control and SMC. It has also been developed in various scenarios such as
multimachine power systems [12], hypersonic vehicles [13], unmanned aerial vehicles [14],
and DC microgrids [15].
However, SMC can only guarantee that the tracking error asymptotically converges to
zero. With respect to this problem, the terminal SMC (TSMC) has attracted more attention,
which can ensure the finite−time convergence of the tracking error, thus improving the
dynamic property of the system [16]. Yao et al. [17] developed a robust adaptive nonsingular
TSMC (NTSMC) in the position and speed control of an automatic train operation system.
Wang et al. [18] adopted continuous TSMC to realize the trajectory tracking of flexible−joint
robots, and Xu et al. [19] introduced a novel nonsingular fast TSMC−reaching law to
shorten the response time of the controlled model.
Although SMC can alleviate the adverse effects of system uncertainty and external
disturbances, it often causes the high−frequency chattering problem due to a large switch-
ing term gain. Moreover, the computation of the control signal depends on exact parameter
information of the system, which is not available in actual application. Some scholars have
developed adaptive sliding mode control (ASMC) [20,21]. In [20], an ASMC was presented
for parameter estimation in a boost converter through the use of state observers. In addition,
Xu and Yao [22] utilized the bounds on the parameters to construct an adaptive projection
algorithm, which ensures that the estimates belong to a known bounded region. However,
when dealing with fast time−varying parameters, control action through ASMC alone is
not reliable. This disadvantage is overcome by employing a neural network feedforward
control method of estimating the nonlinear function.
The neural network (NN) has a strong learning ability, which is widely applied in the
control of unknown systems. In [23], a novel NN−based controller was presented to obtain
the reference tracking control of a nonlinear position servo system, and the neural network
was trained by a Deep Q−Learning algorithm.
The radial basis function neural network (RBFNN), as a common network structure,
is obtained in nonlinear systems with unknown nonlinearities [24]. Moreover, a novel
self−organizing RBFNN was reported in [25], where the structure of the RBFNN can be
regulated according to the system behavior. Considering the low approximation capability
of the RBFNN, and that it takes a long time to adjust weights, this is not a favorable choice.
The fuzzy neural network (FNN) possesses the learning ability of neural networks
and inference capability of fuzzy systems in handling unknown information, which have
been developed rapidly for dynamic modeling and advanced controllers design [26–29].
Lin et al. [30] proposed a Takagi–Sugeno–Kang fuzzy neural network to estimate the
lumped uncertainty. The recurrent neural network (RNN) uses the cyclic connection tech-
nique, and the neuron feedback loop is utilized to transmit previous output signals, which
means that the network considers the previous dynamic information to improve the con-
nection dependency in layers. Therefore, the RNN has the ability to handle time−varying
Mathematics 2023, 11, 1094 3 of 21

inputs and external disturbances. A self−organizing RNN with a general nonlinear form
for predicting system behaviors is proposed in [31]. However, it is difficult for the RNN
with a single hidden layer to maintain high approximation accuracy, which will also in-
crease the calculation computation due to the necessity of a large number of neurons. Thus,
the RNN with a double−loop structure was proposed [32,33]. In [33], this network struc-
ture was designed for identification and prediction of the motion control system. However,
conventional single−layer neural networks may need to train numerous neurons, which
will cause the increase in computational complexity and the excessive consumption of high
memory. Therefore, a multiple−layer neural network (MLNN) structure has been widely
concerned, which can provide fewer neurons training and higher approximation accuracy.
Lee et al. [34] designed a multiple−spatiotemporal−scale recurrent neural network for
action recognition. It was reported in [35] that a convolutional multiple−layer recurrent
neural network can be used for both homogeneous and heterogeneous images. In [36], a
multilayer perceptron neural network was developed as an exploratory test oracle. Some
fuzzy neural controllers were investigated to enhance the approximation capability of
neural networks for dynamic systems [37–43].
Based on the merits of FNN, RNN, and MLNN, this paper designs a multiple−layer
fuzzy neural network structure, i.e., double−hidden−layer recurrent neural network
(DHLRNN), which introduces double hidden layers and double loop feedbacks, suppress-
ing the adverse effects of external disturbances and dealing with the uncertainty of the
nonlinear systems.
Based on the above research, an adaptive backstepping terminal sliding mode control
based on the DHLRNN is proposed in this paper. The DHLRNN and parameter adaptive
law derived from the projection algorithm are utilized to estimate the unknown function
and control coefficient of the universal nonlinear system, respectively, and the estimated
value is then transferred to the controller for effective feedforward compensation. Moreover,
the proposed SMC method combines the features of backstepping control and TSMC; under
the framework of such a control strategy, the system can obtain faster transient responses.
The main contributions of the proposed control method compared with existing works are
summarized as follows:
(1) The proposed DHLRNN inherits the fuzzy inference ability of the FNN, and includes
the self−regulation ability of the RNN, which can be used to estimate more complex
unknown functions. Therefore, the DHLRNN can counteract and compensate system
uncertainties to improve the tracking accuracy of the control system. As an improve-
ment of the single−hidden−layer neural network, the DHLRNN contains two−layer
activation functions for feature extraction. It can process complex data and avoid
serious calculation complexity by reducing the number of neurons.
(2) This paper utilizes Lyapunov stability theory to derive the adaptive laws of the
DHLRNN. The parameters including base width, center vector, and feedback weights
can be trained online. The neural network output can be continuously updated accord-
ing to the system parameter variations. This means that the proposed neural network
will reach the optimal value after a short learning and adjustment. The proposed
controller can obtain dynamic regulation instead of relying on exact mathematical
models. The adaptive law of the control coefficient derived from Lyapunov theory
adopts an expression form of a projection algorithm, which ensures that the estimated
value changes in a known bounded region. Whether faced with disturbances or not,
the control coefficient remains bounded all the time, and the stability of control signal
can be guaranteed.
(3) The design of backstepping TSMC not only reduces the complexity of control design,
but also makes the state tracking error converge in finite time. The use of the switching
term can also counteract the adverse effects of external disturbances and the lumped
approximation error of the neural network to improve the steady−state performance
and anti−disturbance performance of the system.
Mathematics 2023, 11, 1094 4 of 21

The rest of this paper is organized as follows. Section 2 introduces the state equation
of nonlinear systems and the problem statement is presented. In Section 3, the design of
the proposed control method is given, including the neural network definition and stability
proof. The simulation study is shown in Section 4. Finally, Section 5 concludes this paper.

2. System Description and Problem Statement

Consider the following a class of a SISO partially unknown nonlinear system
 .
x 1 = x2
.


 x 2 = x3



.. (1)
.
 .
x = xn

 . n −1



x n = f ( X ) + bu + d(t)

T
where X = [ x1 x2 · · · xn ] is the state vector, and f ( X ) and b are a nonlinear function
and control coefficient, respectively. u denotes the control signal and d(t) denotes the
external disturbance.
Due to the parameter variations and measurement limitations, the system parameters
are difficult to obtain accurately; there exists f ( X ) = f 0 ( X ) + ∆ f ( X ), where f 0 ( X ) is the
nominal part of the nonlinear function and ∆ f ( X ) is the unknown uncertainty. ∆ f ( X ) is
considered upper−bounded. The control coefficient b is considered bounded.
The main control objective is to design an appropriate control signal u so that the state
signal X can accurately track the reference trajectory Xr , i.e.,

lim | X − Xr | = lim | E| = 0 (2)

t→∞ t→∞

. ( n −1) T . T
where Xr = [ xr , xr , · · · , xr ] and E = [e, e, · · · , e(n−1) ] is the error vector.
The BTSMC method is presented for the control problem; f ( X ) and b are assumed to
be known. Moreover, the terminal function p(t) is introduced into the backstepping control
to ensure finite−time convergence, and the sliding surface is built by the backstepping
control variables.

Remark 1. Consider that the terminal function p(t) is a continuous differentiable function with n
order. In order to obtain global robustness, the condition of e(i) (0) = p(i) (0), i = 0, 1, · · · , n − 1 is
.
established. Moreover, for the set time T > 0, when t ≥ T, p(t) = 0, p(t) = 0, · · · , p(n) (t) = 0
should be satisfied to obtain finite−time convergence of the tracking error. Therefore, the terminal
function is defined as
 n n n
(k) k a
 ∑

 1
k! e (0) t + ∑ ∑ [ T j−l +jln+1 e(0)(l ) ] t j+n+1 , i f 0 ≤ t ≤ T
k =0 j =0 l =0
p(t) = (3)


0, i f t > T


.
The terminal function coefficients a jl can be calculated by solving the equations p( T ) = 0, p( T ) = 0,
· · · , p(n) ( T ) = 0.

The detailed design process is described as the following steps

Step 1 Design of the virtual control term α1
(i) Define the tracking error z1 = x1 − xr − p;
(ii) Design the virtual control term α1 = −k1 z1 , where k1 > 0.
Step 2 Design of the virtual control term α2
. .
(i) Define the tracking error z2 = x2 − xr − p − α1 ;
.
(ii) Design the virtual control term α2 = −k2 z2 − z1 + α1 , where k2 > 0;
Mathematics 2023, 11, 1094 5 of 21

.
(iii) Take the derivative of z1 : z1 = −k1 z1 + z2 .
Step i Design of the virtual control term αi
( i −1)
(i) Define the tracking error zi = xi − xr − p ( i −1) − α i −1 ;
.
(ii) Design the virtual control term αi = −k i zi − zi−1 + αi−1 , where k i > 0;
.
(iii) Take the derivative of zi−1 : zi−1 = zi − k i−1 zi−1 − zi−2 .
Step n Design of the control law
( n −1)
(i) Define the tracking error zn = xn − xr − p ( n −1) − α n −1 ;
(ii) Design the following terminal sliding surface s = czn−1 + zn , where c > 0;
.
(iii) Take the derivative of zn−1 : zn−1 = zn − k n−1 zn−1 − zn−2 ;
. (n) .
(iv) Take the derivative of zn : zn = f ( X ) + bu + d(t) − xr − p(n) − αn−1 ;
(v) Design an equivalent control law ueq : ueq = 1b [−c(−k n−1 zn−1 − zn−2 +
(n) .
z n ) − z n −1 − f ( X ) + x r + p ( n ) + α n −1 ] ;
(vi) Design a switching control law usw : usw = 1b [−ρs − λsign(s)], where ρ > 0,
λ > D, and sign(·) denotes a sign function;
(vii) Design the ideal control law u: u = ueq + usw .

Theorem 1. For SISO nonlinear system (1), when all parameter information of the system is known
and the external disturbance d(t) is bounded, if the terminal sliding surface is chosen as

s = czn−1 + zn (4)

and the ideal controller using a switching term is designed as

u= ueq + usw
1 (n) . (5)
= [−c(−k n−1 zn−1 − zn−2 + zn ) − zn−1 − f ( X ) + xr + p(n) + αn−1 − ρs − λsign(s)]
b
all signals of the closed−loop system will asymptotically converge to zero.

Proof. The Lyapunov function is selected as

1 n −1 2 1 2
2 i∑
V= zi + s (6)
=1
2

Taking the first derivative of (6) and combining the relationship between zi and sliding
surface s, it can be obtained that
. n −1 . . n −1 .
V = ∑ zi zi + ss = − ∑ k i z2i + zn−1 zn + ss
i =1 i =1
n −1 (7)
= − ∑ k i z2i + zn−1 zn + s[c(−k n−1 zn−1 − zn−2 + zn )
i =1
(n) .
+ f ( X ) + bu + d(t) − xr − p ( n ) − α n −1 ]

Substituting control input (5) into (7) obtains

. n −1
V = − ∑ k i z2i + zn−1 (s − czn−1 ) + s(−zn−1 − ρs − λsign(s) + d(t))
i =1
n −1
= − ∑ k i z2i − cz2n−1 − ρs2 − λ|s| + sd(t) (8)
i =1
n −1
≤ − ∑ k i z2i − cz2n−1 − ρs2 − λ|s| + D |s|
i =1
Mathematics 2023, 11, 1094 6 of 21

Assume |d(t)| ≤ D, where D is an unknown positive constant. When the condition

λ > D is satisfied, the following inequality holds:

. n −1
V ≤ − ∑ k i z2i − cz2n−1 − ρs2 − (λ − D )|s|
i =1 (9)
n −1
≤−∑ k i z2i − cz2n−1 − ρs2
i =1
.
Therefore, inequality (9) satisfies V ≤ 0. According to the LaSalle invariance principle,
the convergence of backstepping control variables and the sliding surface can be guaranteed,
i.e., when t → ∞ , then zi = 0 (i = 1, 2, · · · n), s = 0. Moreover, the tracking error E will
converge to zero in finite time under the constraint of terminal function p(t). The detailed
proof is shown in Remark 5. 

Remark 2. When the Lyapunov function and system control law are selected, then the condition
. . .
of lim ss ≤ 0 is derived, and it can be concluded that lim s ≤ 0 and lim s ≥ 0. This means that
s →0 s →0+ s →0−
as long as there are state signals close to switching surface s( x ) = 0, they will be attracted and
fluctuate around the switching surface. Therefore, according to the designed sliding mode trajectory,
the state signals can reach the specified region.

3. Design of Adaptive Backstepping Terminal Sliding Mode Controller Using Dhlrnn

Although the ideal control signal designed in (5) can guarantee the convergence of all
states and signals of the nonlinear system (1), an accurate mathematical model is difficult
to obtain because of environmental fluctuations and disturbances. Control actions through
SMC alone result in a large switching term gain, which will cause system chattering and
damage the steady−state performance of the system. In order to observe the real−time
information of the model, this section utilizes the DHLRNN and parameter adaptive laws
to estimate the nonlinear function f ( X ) and control coefficient b, respectively, and the
estimated value is introduced into the controller for effective feedforward compensation to
improve control accuracy.

3.1. Structure of DHLRNN

The specific structure of the proposed neural network is shown in Figure 1, which
comprises the input layer (layer 1), the first hidden layer (layer 2), the second layer (layer 3),
and the output layer (layer 4).
The proposed DHLRNN is a multi−layer network structure and possesses self−learning
ability, which combines the advantages of the FNN and RNN. It can be used to estimate
complex nonlinear functions, which provides an effective solution to the unknown uncer-
tainty problem of the system.
The detailed signal transmission and the basic function in each layer of the DHLRNN
are introduced as follows
Layer 1 (Input Layer): The main function of this layer is to pass the input variables to
the next layer. Due to the design of external feedback, the output value exY at the previous
time is connected to the input layer by the feedback weight Wroi (i = 1, 2, · · · , m), thus
providing the optimal input for the subsequent processing. For every node in the input
layer, the ith node output is calculated as follows:

θi = xi · Wroi · exY, (i = 1, 2, · · · , m) (10)

where xi denotes the input of the ith node, X = [ x1 , x2 , · · · , xm ] T is the input vector in this
layer, and θ = [θ1 , θ2 , · · · , θm ] T is the output vector. The external feedback weight matrix of
the input layer is described as

Wro = [Wro1 , Wro2 , · · · , Wrom ] T (11)

 forward compensation to improve control accuracy.

3.1. Structure of DHLRNN

The specific structure of the proposed neural network is shown in Figure 1, which
Mathematics 2023, 11, 1094 comprises the input layer (layer 1), the first hidden layer (layer 2), the second layer7 (lay-
of 21

er 3), and the output layer (layer 4).

 Output layer

W1 W2 Wl
Z−1 Z−1
The second
h21 h22  h2 l hidden layer

Z−1 Z−1 Z−1 Z−1 The first

h11 h12 h13  h1n hidden layer
Wr1 Wr 2 Wr 3 Wrn

Wro1 Wrom

θ1  θm Input layer

x1 xm

Figure1.1.Structure
Figure Structureof
ofthe
theproposed
proposedneural
neuralnetwork.
network.

Layer
The 2proposed
(The First hidden
DHLRNNlayer):isThis
a layer adopts anetwork
multi−layer Gaussianstructure
function asand
an activation
possesses
function to realize
self−learning fuzzification
ability, operationthe
which combines with the aim to expand
advantages the dimension
of the FNN and RNN. of the input
It can be
variables. The signal
used to estimate is mapped
complex from the
nonlinear low−dimensional
functions, which provides space toan high−dimensional
theeffective solution to
space. The internal
the unknown feedback
uncertainty loop is
problem ofrealized
the system. and the dynamic recursion leads to better
mapping, which improves the neuron activation
The detailed signal transmission and the basic intensity. The output
function in eachform of of
layer jth DHL-
thethe node
isRNN
expressed as
are introduced as follows
−net1j
1j = efunction
Layer 1 (Input Layer): Theφmain , ( j of
= 1, this2, ·layer
· · , nis
) to pass the input variables to
m (θ ·W · Lφ −c )2
the next layer. Due to the design of external
net1j = ∑ i feedback,
rj 1j 1j the output value exY at the (12) pre-
2
b1j
vious time is connected to the input layer by the feedback weight Wroi (i = 1, 2,, m) ,
i = 1

where b1j and c1jthe

thus providing areoptimal
the center vector
input andsubsequent
for the the base width of the jth
processing. activation
For every node function,
in the
input layer, the 1jith node output is calculated as follows:
respectively. Lφ represents the previous output of the jth node, which will be fed back
to the current node through the feedback weight Wrj . The output vector of the first
hidden layer is Φ1 = [φ11 , φ12 ,θ· · ·=, x ] T . The
φ1n⋅W
roi ⋅ exY ,(i = 1,2,, m)
center vector matrix, base width matrix, and
(10)
i i
feedback weight matrix of the first hidden layer are, respectively, described as

where xi denotes the input of B the=ith T

, b12 , · · ·X, b=1n[ x]1T, x 2 ,  , x m ] is the input vector
[b11 node, in
(13)
1
this layer, and θ = [θ ,θ
1 2 ,,θ m ]T
is the output vector. The external feedback weight matrix
T
of the input layer is described asC1 = [c11 , c12 , · · · , c1n ] (14)

r =
WW [Wr1 , Wr2 , · · · , Wrn ] T T
ro = [W ro 1 , W ro 2 ,  , W rom ]
(15)
(11)
Layer
Layer32(The
(Thesecond
First hidden
hiddenlayer):
layer):This
Thislayer
layeralso utilizes
adopts Gaussian function
a Gaussian function for nonlinear
as an activa-
mapping, which is similar to the function of the first hidden layer. The signals are mapped
tion function to realize fuzzification operation with the aim to expand the dimension of
from the first hidden layer space to the higher−dimensional space, and the features of
the input vector are further extracted. The kth node output of the second hidden layer is
expressed as
φ2k = e−net2k , (k = 1, 2, · · · , l )
n ( φ − c )2
net2k = ∑ 1j b2 2k (16)
j =1 2k

where b2k and c2k denote the center vector and the base width, respectively. Φ2 = [φ21 , φ22 , · · · , φ2l ] T
is the output vector of the second hidden layer. The center vector matrix and base width matrix of the
second hidden layer are, respectively, described as

B2 = [b21 , b22 , · · · , b2l ] T (17)

Mathematics 2023, 11, 1094 8 of 21

C2 = [c21 , c22 , · · · , c2l ] T (18)

Layer 4 (Output layer): This layer will calculate the overall output of the network and
serve as the feedback signal of the input layer. The single output node is connected with
each neuron of the second hidden layer by the weight Wk (k = 1, 2, · · · , l ), which is labeled
with the symbol Σ. The output of the DHLRNN is the weighted sum of the Gaussian
function calculated by the second hidden layer, expressed as

l
Y = W T Φ2 = ∑ Wk φ2k (19)
k =1

where the weight matrix of the output layer is described as

W = [W1 , W2 , · · · , Wl ] T (20)

Based on the number of nodes m, n, l in each layer, the proposed DHLRNN structure
is defined as m − n − l − 1, which can be selected carefully according to the specific system.

3.2. Approximation Error of DHLRNN

The DHLRNN is used to estimate the nonlinear function in the ideal controller (5),
and there exists B1∗ , C1∗ , Wr∗ , B2∗ , C2∗ , Wro
∗ , and W ∗ , satisfying the following expression

T
f ( X ) = W ∗ Φ2∗ ( B1∗ , C1∗ , Wr∗ , B2∗ , C2∗ , Wro
∗
)+ε (21)

where ε is the reconstruction error between the optimal value and actual value, bounded
as: ε ≤ ε N , and ε N is a positive constant.
The optimal parameters of the DHLRNN are obtained from the following equation

B1∗ , C1∗ , Wr∗ , B2∗ , C2∗ , Wro

∗ , W∗

= argmin [supk f − f ∗ k] (22)

( B1 ,C1 ,Wr ,B2 ,C2 ,Wro ,W )

The estimated value of f ( X ) is the real output of the neural network, expressed as

fˆ( X ) = Ŵ T Φ̂2 ( B̂1 , Ĉ1 , Ŵr , B̂2 , Ĉ2 , Ŵro ) (23)

where Ŵ, B̂1 , Ĉ1 , Ŵr , B̂2 , Ĉ2 , Ŵro are the estimated values of the neural network parameters.
Therefore, the approximation error between the estimated value and actual value is
defined as
T
f − fˆ = W ∗ Φ ∗
 2 − Ŵ Φ̂
T
2 + ε
∗ T
=W Φ̂2 + Φe 2 − Ŵ T Φ̂2 + ε
∗T ∗T e
 Φ̂2 + W Φ2 − Ŵ Φ̂2 + ε
=W T (24)
T
= Ŵ + W e Φ̂2 + Ŵ + W
T T eT Φ
e 2 − Ŵ T Φ̂2 + ε
=We T Φ̂2 + Ŵ T Φ
e 2 + ε0

where ε 0 = W e TΦe 2 + ε denotes the total approximation error and W e = W ∗ − Ŵ is the

weight error.
In order to convert nonlinear f ( X ) into a partially linear form, the Taylor expansion
linearization technique is used in this paper. The Taylor expansion of the optimal output
Mathematics 2023, 11, 1094 9 of 21

vector Φ2∗ of the second hidden layer at B1∗ = B̂1 , C1∗ = Ĉ1 , Wr∗ = Ŵr , B2∗ = B̂2 , C2∗ = Ĉ2 ,
and Wro∗ = Ŵ can be calculated as follows
ro

Φ2∗ ( B1∗ , C1∗ , Wr∗ , B2∗ , C2∗ , Wro

∗ ) = Φ̂ ( B̂ , Ĉ , Ŵ , B̂ , Ĉ , Ŵ )
2 1 1 r 2 2 ro
+ ∂Φ
∂B∗
2
( B1∗ − B̂1 ) +
∂Φ2 ∗
∂C1∗ C ∗ =Ĉ (C1 − Ĉ1 )
1 B1∗ = B̂1 1 1
∂Φ2 ∗ − Ŵ ) + ∂Φ2 ∗ (25)
+ ∗
∂Wr W ∗ =Ŵ ( W r r ∂B2 B∗ = B̂ ( B2 − B̂2 )
∗
r r 2 2
∂Φ2 ∗ ∂Φ2 ∗ − Ŵ ) + O
+ ∂C2∗ C ∗ =Ĉ (C2 − Ĉ2 ) + ∂Wro ∗
∗ =Ŵ
Wro
(Wro ro h
2 2 ro

Φ
e 2 (B
e1 , C
e1 , W
e r, B
e2 , C e ro ) = Φ2B · B
e2 , W
1
e1 + Φ2C · Ce1
1 (26)
+Φ2Wr · Wr + Φ2B2 · B2 + Φ2C2 · C2 + Φ2Wro · Wro + Oh
e e e e

where Oh is a high−order term of expansion; Φ2B1 , Φ2C1 , Φ2Wr , Φ2B2 , Φ2C2 , and Φ2Wro
are first−order partial derivatives of Φ2 to B1∗ , C1∗ , Wr∗ , B2∗ , C2∗ , and Wro
∗ , respectively.

These partial derivatives conform to the arrangement of the Jacobian matrix, which can be
expressed, respectively, as the following forms:
 ∂φ ∂φ21 
21
∂b11 ··· ∂b1n
 . .. .. 
Φ2B1 = 
 .. . . 
 (27)
∂φ2l ∂φ2l
···
∂b11 ∂b1n l ×n
 ∂φ ∂φ21 
21
∂c11 ··· ∂c1n
 . .. .. 
Φ2C1 = 
 .. . . 
 (28)
∂φ2l ∂φ2l
···
∂c11 ∂c1n l ×n
 ∂φ ∂φ21 
21
∂W ··· ∂Wrn
 . r1 .. .. 
Φ2Wr  ..
= . . 
 (29)
∂φ2l ∂φ2l
···
∂Wr1 ∂Wrn l ×n
 ∂φ ∂φ21 
21
∂b ··· ∂b2l
 .21 .. .. 
Φ2B2  ..
= . . 
 (30)
∂φ2l ∂φ2l
···
∂b21 ∂b2l l ×l
 ∂φ ∂φ21 
21
∂c21 ··· ∂c2l
 . .. .. 
Φ2C2  ..
= . . 
 (31)
∂φ2l ∂φ2l
···
∂c21 ∂c2l l ×l
 ∂φ21 ∂φ21 
∂Wro1 ··· ∂Wrom
 . .. ..
Φ2Wro

 ..
= . . (32)


∂φ2l ∂φ2l
···
∂Wro1 ∂Wrom l ×m
Substituting partial derivatives (27)–(32) into (24), the approximation error can be
further expressed as

f ( X ) − fˆ( X ) =We T Φ̂2 + Ŵ T Φ

e 2 + ε0
=We T Φ̂2 + Ŵ T Φ2C C e + Ŵ T Φ2C C
1 1 2 2
e
(33)
+Ŵ Φ2B1 B1 + Ŵ Φ2B2 B2 + Ŵ Φ2Wr W
T e T e T er
+Ŵ Φ2Wro W
T e ro + ∆0

where ∆0 = Ŵ T Oh + ε o denotes the lumped high−order approximation error.

Mathematics 2023, 11, 1094 10 of 21

Assumption 1. Due to the approximation capability of the DHLRNN, the lumped high−order
approximation error of the DHLRNN, i.e., ∆o , is much smaller than the function being estimated,
i.e., f ( X ). ∆o is bounded and satisfies |∆o | ≤ ∆d , where ∆d is a positive constant.

3.3. Stability Analysis

The block diagram of ABTSMC based on the proposed DHLRNN is shown in Figure 2.
The neural network parameters are updated online through the adaptive laws, so the
output of the neural network can adaptively adjust to the optimal value. The flowchart of
the learning algorithm of the proposed controller is provided in Figure 3, which comprises
the detailed algorithm description. Furthermore, an additional parameter adaptive law is
Mathematics 2023, 11, x FOR PEER REVIEW 12 of 24
REVIEW to estimate the control coefficient b for guaranteeing the stability of the control
Mathematics 2023, 11, x FOR PEER adopted 12law
of 24
u, and the corresponding estimated value is denoted as b̂.

Terminal
Terminal
function P(t)
function P(t)

R E Backstepping z1 , z2 ,, zn
R E Backstepping z1 , z2 ,, zn Terminal Terminal
− control sliding surface
− control sliding surface External
+ External
disturbance
+ disturbance
X ABTSMC with
X ABTSMC with X = f ( X ) + bU + d (t )
DHLRNN
DHLRNN
X = f ( X ) + bU + d (t )

bˆ = Proj ˆ (η8 su )
bˆ =b Projbˆ (η8 su )
W = −η1 sΦ ˆ fˆ ( X )ˆ= Wˆ T Φ
ˆ
2ˆ T ˆ
W = −η1 sΦ
2
ˆ b̂ f (X ) = W Φ2
b̂
B1T = −ηT 2 sWˆ T Φ 2 B1 T
2

B1 = −η2 sWˆ Φ 2 B1

 T ˆ
C1 = −ηT 3 sW Φ 2C1 T
T

C1 = −η3 sWˆ Φ 2C1 E DHLRNN

WrT = −ηT 4 sWˆ T Φ 2Wr T
E approximator
DHLRNN
Wr = −η4 sWˆ Φ 2Wr approximator

B 2T = −ηT 5 sWˆ T Φ 2 B2 T
B 2 = −η5 sWˆ Φ 2 B2

C 2T = −ηT 6 sWˆ T Φ 2C2 T
C 2 = −η6 sWˆ Φ 2C2
WroT = −ηT 7 sWˆ T Φ 2WroT
Wro = −η7 sWˆ Φ 2Wro
Adaptive laws
Adaptive laws

Blockdiagram
Figure2.2.Block
Figure diagramofofthe
theABTSMC
ABTSMCusing
usingDHLRNN.
DHLRNN.
Figure 2. Block diagram of the ABTSMC using DHLRNN.

Start
Start
Definition of Backstepping Variables
Definition of Backstepping Variables
z1 , z2 , , zn
z1 , z2 , , zn
Calculation of Sliding Surface
Calculation of Sliding Surface
s = czn −1 + zn
s = czn −1 + zn
Approximation Strategy Calculation of Control Input Signal
Approximation Strategy Calculation of Control Input Signal
The overall output of the DHLRNN is utilized 1
The overall
to estimate nonlinear output of the DHLRNN
function f ( X ) , i.e., is utilized u = [−c1(− kn −1 zn −1 − zn − 2 + zn ) − zn −1 − fˆ ( X )ˆ
to estimate nonlinear function f ( X ) , i.e., bˆu = ˆ [−c(− kn −1 zn −1 − zn − 2 + zn ) − zn −1 − f ( X )
fˆ ( X )ˆ= Wˆ T Φˆ ( Bˆ , Cˆ ,Wˆ , Bˆ , Cˆ , Wˆ ) b
rˆ 2ˆ 2ˆ ˆ ˆ + xr( n ) + (pn )( n ) + α(nn) −1 − ρ s − λ sign( s )]
f ( X ) = W2ˆ T Φ
ˆ1 (1Bˆ , C
2 1
ro
1 , Wr , B2 , C2 , Wro ) + xr + p + α n −1 − ρ s − λ sign( s )]
The projection algorithm is utilized to estimate
Thecoefficient
control projectionbalgorithm is utilized to estimate
control coefficient b Update Learning Rates
Update
η1 ,η 2 ,Learning
,η8 Rates
η1 ,η 2 , ,η8
Online Parameters Learning
Online Parameters Learning
 
W = −η1 sΦ ˆ B 2T = −Tη5 sWˆ T Φ 2ˆBT2 Does the neural network converge No
W = −η1 sΦ 2 ˆ B2 = −η5 sW Φ 2 B2 Does theoptimal
neural network No
2
value? converge
B1T = −Tη 2 sWˆ T Φ 2ˆBT1  to the
C T = −η sWˆ T Φ
B1 = −η 2 sW Φ 2 B1 2 C 2T 6= −η6 sW2ˆCT2 Φ 2C
to the optimal value?

C1T = −Tη3 sWˆ T Φ 2ˆCT1 W T = −η sWˆ T Φ
2

C1 = −η3 sW Φ 2C1 ro WroT 7= −η7 sW2ˆWTroΦ 2W Yes

 
WrT = −Tη 4 sWˆ T Φ 2ˆWTr
ro

bˆ = Proj Yes
Wr = −η 4 sW Φ 2Wr  ˆ (η8 su )
bˆ =bProj bˆ
(η8 su ) End
End

Figure 3. The flowchart of learning algorithm of the proposed controller.

Figure
Figure 3. 3.
TheThe flowchart
flowchart of of learning
learning algorithm
algorithm of of
thethe proposed
proposed controller.
controller.
As the DHLRNN and parameter adaptive law introduce the estimated information
As the DHLRNN and parameter adaptive law introduce the estimated information
of the system into the controller, the control system can maintain robustness in rejecting
of the system into the controller, the control system can maintain robustness in rejecting
uncertainties and disturbances. The proposed control method no longer depends on the
uncertainties and disturbances. The proposed control method no longer depends on the
accurate mathematical model, which is more suitable for an actual nonlinear system.
accurate mathematical model, which is more suitable for an actual nonlinear system.
Mathematics 2023, 11, 1094 11 of 21

As the DHLRNN and parameter adaptive law introduce the estimated information
of the system into the controller, the control system can maintain robustness in rejecting
uncertainties and disturbances. The proposed control method no longer depends on the
accurate mathematical model, which is more suitable for an actual nonlinear system.
Based on the designed feedforward compensation technology, the new control law is
designed as
u = 1 [−c(−k n−1 zn−1 − zn−2 + zn ) − zn−1 − fˆ( X )
b̂ (34)
(n) .
+ xr + p(n) + αn−1 − ρs − λsign(s)]

Theorem 2. For SISO nonlinear system (1), when the external disturbance d(t) and control
coefficient b are both bounded, if the control law of system is chosen as (34), then it guarantees that
all signals of the closed−loop system will asymptotically converge to zero.

Proof. The new Lyapunov function is selected as

n −1
1
V= 2 ∑ z2i + 12 s2 + 1 eT e
2η1 W W + 1 eT e
2η2 B1 B1 + 1 eT e
2η3 C1 C1
i =1 (35)
+ 2η14 W
e rT W
er + 1 eT e
2η5 B2 B2 + 1 eT e
2η6 C2 C2 + 1 eT e
2η7 Wro Wro + 1 e2
2η8 b

where e b = b − b̂ is the approximation error of the control coefficient; η1 , η2 , η3 , η4 , η5 , η6 ,

η7 , and η8 are the learning rates parameters. 

Remark 3. According to the main control objective, in order to ensure that the state signal can
track the reference trajectory, the convergence of backstepping intermediate variables, the sliding
surface, and the estimated term must be guaranteed. Specially, the seven adaptive parameters of
the DHLRNN including center vectors, base widths, output weight, and feedback weight should
converge to the optimal value so that the neural network can approximate the real−time variation in
the nonlinear function accurately. Motivated by the abovementioned ideas, the Lyapunov function
should contain three parts: backstepping intermediate variables zi (i = 1, 2, · · · , n − 1), sliding
surface s, and estimation error of all parameters. Moreover, according to the Lyapunov stability
theory, we must guarantee that the designed Lyapunov function is positive definite and its derivative
is semi−negative definite. Therefore, the new Lyapunov function is constructed as provided in (35).

Taking the first derivative of the Lyapunov function (35), and then substituting the
relationship between zi (i = 1, 2, · · · , n) and sliding surface s into the derivative, yields

. n −1 . . n −1 .
V = ∑ zi zi + ss + H = − ∑ k i z2i + zn−1 zn + ss + H
i =1 i =1
n −1 (36)
= − ∑ k i z2i + zn−1 zn + s[c(−k n−1 zn−1 − zn−2 + zn )
i =1
(n) .
+ f ( X ) + bu + d(t) − xr − p ( n ) − α n −1 ] + H

where
. .T .T . T
1 eT e 1 e e 1 e e 1 e e
H= η1 W W + η2 B1 B1 + η3 C1 C1 + η4 W r Wr
.T .T . T . (37)
+ η15 B
e2 Be2 + 1 C
η6
e2 + 1 W
e2 C
η7
e ro − 1 e
e ro W
η8 b b̂

Substituting the proposed control law (34) into (37) obtains

. n −1
V= − ∑ k i z2i + zn−1 (s − czn−1 ) + s(−zn−1 + f ( X )
i =1 (38)
− fˆ( X ) + ebu − ρs − λsign(s) + d(t)) + H
Mathematics 2023, 11, 1094 12 of 21

Because (38) contains the approximation error of the function being estimated, bringing
.
expression (33) into (38), V can be further derived as

. n −1
V= e T Φ̂2 s + Ŵ T Φ2B B
− ∑ k i z2i − cz2n−1 + ebsu + W es
1 1
i =1
+Ŵ T Φ2C1 Ce1 s + Ŵ T Φ2Wr W
e r s + Ŵ T Φ2B B
es
2 2
(39)
+Ŵ Φ2C2 C2 s + Ŵ Φ2Wro Wro s − ρs − λ|s|
T e T e 2

+s(∆o + d(t)) + H

The following adaptive laws of the DHLRNN are designed:

.
Defining We T Φ̂2 s + 1 W
e TW
e = 0, it can be obtained that
η1

.
e = −η1 sΦ̂2
W (40)
.T
Defining Ŵ T Φ2B1 B
e1 s + 1 e e
η2 1 B1
B = 0, it can be obtained that

.T
e1 = −η2 sŴ T Φ2B
B (41)
1

.T
Defining Ŵ T Φ2C1 C
e1 s + 1 e e
η3 1 C1
C = 0, it can be obtained that

.T
e1 = −η3 sŴ T Φ2C
C (42)
1

. T
Defining Ŵ T Φ2Wr W
e rs + 1 e e
η4 W r Wr = 0, it can be obtained that

. T
e r = −η4 sŴ T Φ2W
W (43)
r

.T
Defining Ŵ T Φ2B2 B
e2 s + 1 e e
η5 B2 B2 = 0, it can be obtained that

.T
e2 = −η5 sŴ T Φ2B
B (44)
2

.T
Defining Ŵ T Φ2C2 C
e2 s + 1 e e
η6 2 C2
C = 0, it can be obtained that

.T
e2 = −η6 sŴ T Φ2C
C (45)
2

. T
Defining Ŵ T Φ2Wro W
e ro s + 1 e
η7 W ro Wro
e = 0, it can be obtained that

. T
e ro = −η7 sŴ T Φ2W
W (46)
ro

The parameter adaptive law of the control coefficient is designed as follows, which
utilizes the calculation form of the adaptive projection algorithm
.
b̂ = Projb̂ (η8 su) (47)
Mathematics 2023, 11, 1094 13 of 21

Remark 4. The adaptive projection algorithm [20] can be defined as follows:


0 b̂ ≥ bl and · > 0
Projb̂ (·) = 0 b̂ ≤ bo and · < 0 (48)
· otherwise


.
When b̂ exceeds the maximum bl and continues to increase, then b̂ = 0, which means that b̂ will
remain unchanged. When b̂ is less than the minimum bo and continues to decrease, b̂ will remain
unchanged. In other cases, the adaptive law acts normally. To sum up, this algorithm can ensure the
boundedness of the estimated term and avoid the sudden variation in b̂ with different disturbances
so that the control signal of the system can achieve stable control performance.
.
Substituting the above adaptive laws (40)–(47) into (39), V is simplified as

. n −1
V ≤ − ∑ k i z2i − cz2n−1 − ρs2 − λ|s| + s(∆o + d(t)) (49)
i =1

Because |d(t)| ≤ D and |∆o | ≤ ∆d , then Equation (49) can be further expressed as

. n −1
V ≤ − ∑ k i z2i − cz2n−1 − ρs2 − λ|s| + s( D + ∆d ) (50)
i =1

By selecting the switching term gain, as long as λ satisfies the condition of λ > D + ∆d ,
the following result can be concluded:

. n −1
V ≤ − ∑ k i z2i − cz2n−1 − ρs2 − (λ − D − ∆d )|s|
i =1
n −1 (51)
≤ − ∑ k i z2i − cz2n−1 − ρs2
i =1
≤0
. .
Because V ≤ 0 guarantees that V is semi−negative definite, according to the LaSalle
invariance principle, when t → ∞ , zi = 0(i = 1, 2, · · · , n) and s = 0.
Meanwhile, the finite−time convergence of tracking error E is illustrated in Remark 5.

Remark 5. According to Remark 1 and the expression of p(t), when t = 0, e(i) (0) = p(i) (0) (i =
1, 2, · · · , n − 1). Therefore, it can be obtained that

z1 (0) = e (0) − p (0) = 0

. .
z2 (0) = e (0) − p (0) − α1 (0) = 0
.. ..
z3 (0) = e (0) − p (0) − α2 (0) = 0
.. (52)
.
z n (0 ) = e ( n −1) (0 ) − p ( n −1) (0 ) − α n −1 (0 ) = 0
s(0) = czn−1 (0) + zn (0) = 0

It can be seen from (52) that all initial states of the system are already equal to zero, and it has
been proven that lim zi = 0 and lim s = 0, so the reaching condition of the BTSMC is eliminated.
t→∞ t→∞
This means that the condition of zi = 0 (i = 1, 2, · · · , n), s = 0 always holds; thus, it is deduced
that αi = 0 (i = 1, 2, · · · , n − 1) is always satisfied, and the global robustness of closed−loop
system is guaranteed.
.
Because the terminal function has ensured that p(t) = 0, p(t) = 0, · · · , p(n−1) (t) = 0 for
t ≥ T, the tracking error E can be guaranteed to converge to zero in finite time.
 i

deduced that α i = 0 (i = 1, 2, , n − 1) is always satisfied, and the global robustness of

closed−loop system is guaranteed.
Because the terminal function has ensured that p (t ) = 0 , p (t ) = 0 ,  , p ( n −1) (t ) = 0 for
t ≥ T , the tracking error E can be guaranteed to converge to zero in finite time.
Mathematics 2023, 11, 1094 14 of 21

Remark 6. The unknown function f ( X ) possesses a strong nonlinear characteristic; when

faced with the inevitable system uncertainties and disturbances, its value will experience unex-
Remark 6. The unknown
pected variation. Thus, thefunction f ( X )law
ideal control possesses a strong
(5) with nonlinear term
the uncertainty characteristic; when faced
has a high−frequency
with the inevitable system uncertainties and disturbances, its value will experience
fluctuation. In order to weaken the chattering phenomenon, the proposed controller utilizes the unexpected
variation. Thus, the ideal control law (5) with the uncertainty term has a high−frequency fluctuation.
DHLRNN to estimate f ( X ) for accurate feedforward compensation to observe the real−time
In order to weaken the chattering phenomenon, the proposed controller utilizes the DHLRNN to
estimate ( Xaccurate
( X ) ffor
variationf in ) . The signum function
feedforward
sign()
compensationistoadopted
observe to
thecompensate the approximation
real−time variation in f ( X ).
error.
The As long
signum as an accurate
function sign(·) isapproximation of the uncertainty
adopted to compensate term is obtained,
the approximation thelong
error. As switching
as an
term gain is set to be small. Therefore, the discontinuous behavior of the controller
accurate approximation of the uncertainty term is obtained, the switching term gain is set to can be effec-
be
tively suppressed and the chattering is reduced.
small. Therefore, the discontinuous behavior of the controller can be effectively suppressed and the
chattering is reduced.
4. Simulation Study
In this section,
4. Simulation Studya DC–DC buck converter model is built to verify the effectiveness of
the proposed control
In this section, methodbuck
a DC–DC (ABTSMC−DHLRNN). Theto
converter model is built basic circuit
verify framework of
the effectiveness of the
the
model is shown
proposed control in Figure(ABTSMC
method 4, which−comprises
DHLRNN). a load circuit Rframework
resistance
The basic 0 , a LC filter,
of theamodel
diode
isDshown
, a PWM gate drive−controlled
in Figure switch
4, which comprises (insulated
a load gate
resistance R0bipolar a diodeQD,, and
transistor)
, a LC filter, a PWM DC
gate driveV−
voltage controlled switch (insulated gate bipolar transistor) Q, and DC voltage V lines
in . The controlled switch ON and OFF cases are depicted with dashed in
. The1
controlled switch
and 2, respectively. ON and OFF cases are depicted with dashed lines 1 and 2, respectively.

u PWM

Q
L

iL
iC iR
+
+
Vin ① D ② C R0 vo
−
−

Figure 4.
Figure 4. Circuit
Circuit diagram
diagram of
of DC–DC
DC–DCbuck
buckconverter
converter(ON
(ONcase:
case: Line
Line1,
1,OFF
OFFcase:
case: Line
Line 2).
2).

The Kirchhoff law and state space method are utilized here to analyze the circuit
model. The dynamic model can be built along the following steps, where vo and i L denote
output voltage and inductor current, respectively.
When controlled switch Q is turned on, vo and i L satisfy

 C dvo = i L − vo

dt R0
di (53)
 L L = Vin − vo

dt
When controlled switch Q is turned off, vo and i L satisfy

 C dvo = i L − vo

dt R0
di (54)
L
 L

= −vo
dt
Therefore, the dynamic model of the DC–DC buck converter can be written as

 di L = − 1 vo + Vin u

dt L L
dv (55)
o 1 1

 = C i L − R0 C v o
dt
Mathematics 2023, 11, 1094 15 of 21

Then, the following second−order state equation is designed

 .
x 1 = x2
. (56)
x2 = f ( x ) + bu + d(t)
.
where x1 = vo and x2 = vo represent state variables, f ( x ) = − LC x1
− Rx02C , and b = VLC
in
.
Based on the design in Section 3, the control law (34) is applied to the simulation model.
The reference voltage is denoted as vre f . The control objective of the buck converter is
to design a control law, allowing the output voltage to quickly track the reference voltage
trajectory. According to the state equation of the buck converter system, the ABTSMC with
the DHLRNN is designed as

u= 1
b̂..
(−c(−k1 z1 + z2 ) − z1 − fˆ( x )
.. . (57)
+vre f + p − k1 z1 − ρs − λsgn(s))

The nominal parameter values of the main circuit are summarized in Table 1. The
switching frequency for IGBT is chosen as 10 kHz and the simulation sampling period is
chosen as 10 µs.

Table 1. Model nominal parameters.

Description Parameter Value Units

Input voltage Vin 25 V
Reference voltage vre f 12 V
Inductor L 6 mH
Capacitor C 2200 µF
Load resistance R0 30 Ω
Switching frequency f sw 10 kHz
Sampling period Ts 10 µs

To illustrate the superiority of the proposed control method, the ABTSMC control
method without neural network approximation (ABTSMC) and BSMC based on the RBFNN
(BSMC−RBFNN) are investigated here for comparison. The above controllers are simulated
on a digital simulation platform using Matlab/Simulink software.
For fair comparison, the three control strategies are carried out under identical condi-
tions to the simulation study, and the selection of the sliding mode gain is the same. The
parameters of the above controllers are selected as listed in Table 2.

Table 2. Controller parameters.

Controllers Parameters and Values

ABTSMC c = 4000, k1 = 2 × 105 , ρ = 2000, λ = 1200, T = 0.01
BSMC−RBFNN c = 4000, k1 = 2 × 105 , ρ = 100, λ = 120
ABTSMC−DHLRNN c = 4000, k1 = 2 × 105 , ρ = 2000, λ = 1200, T = 0.01

Remark 7. The switching term gain existing in SMC affects the steady−state performance of the
system. Due to the discontinuity of the signum function, too large a switching gain often brings
the system chattering problem. Therefore, in order to highlight the neural network approximation,
the switching gain is set to be small. The relevant sliding gain is set to be large for obtaining fast
convergence. Moreover, with regard to the learning rates of the neural network, each parameter can
be observed according to the response curve; thus, we can judge whether these parameters converge.
Among them, output weight determines the output signal amplitude of the neural network; in order
to cope with the high amplitude of the nonlinear function, the learning rate of the weight is set larger
than the other parameters. When all the learning rates are set to the appropriate order of magnitude,
 Mathematics 2023, 11, 1094 16 of 21

the output signal can roughly track the reference trajectory. Then, they can be fine−tuned until the
control objective is achieved.

To compare the performance of the above controllers under different uncertainties and
disturbances, the buck converter is subjected to the following four situations: (1) start−up
phase analysis, (2) load resistance variations, (3) reference voltage variations, and (4) input
voltage variations.
Mathematics 2023, 11, x FOR PEER REVIEW In addition, the structure of the DHLRNN is selected as 2−4−3−1, and the learning 19 of 24
rates in (40)–(47) are η1 = 2.6 × 105 , η2 = 0.02, η3 = 0.13, η4 = 0.1, η5 = 0.01, η6 = 2 × 10−5 ,
η7 = 1 × 10−10 , and η8 = 16000.

BSMC−RBFNN method
4.1. Start−Up Phase produces a larger output voltage overshoot during the start−up
Analysis
phase.TheFurthermore,
reference voltageto
due thetolack
is set 12 V,ofand
terminal
the loadfunction, the
resistance tracking
remains error will
unchanged notΩ.
at 30
converge in finite
The response time.obtained
curves Therefore,forthe
thecorresponding
output voltagestartup time is current
and inductor longer than those ofin
are shown
other control strategies, which takes about 260 ms. Although the ABTSMC
Figure 5a,b, respectively. As shown in Figure 5a, due to the low compensation accuracy method can
reduce the convergence time via introducing the terminal function, the
of the RBFNN, a satisfactory dynamic response is difficult to obtain. The BSMC−RBFNN problem of large
output
method voltage overshoot
produces a largeralso existsvoltage
output due toovershoot
the lack ofduring
an estimation
the startitem
−upof the nonline-
phase. Further-
armore,
function in the controller.
due to the lack of terminal function, the tracking error will not converge in finite time.
On the the
Therefore, contrary, the proposed
corresponding startup control method
time is longertracks the reference
than those of other voltage in 10 ms
control strategies,
without overshoot.
which takes about The behavior
260 ms. of inductor
Although currentmethod
the ABTSMC as shown caninreduce
Figurethe
4b also further
convergence
illustrates the superiority
time via introducing of the proposed
the terminal function, control method.
the problem of large output voltage overshoot
also exists due to the lack of an estimation item of the nonlinear function in the controller.

260ms
130ms
10ms

0.8V 6.9A
5.2A
1.2V 3A
10ms
30ms
30ms

(a) (b)

Figure
Figure5.5.Response
Responsecurves
curvesofofstart−up
start−upphase.
phase.(a)(a)
Output voltagevovo, ,V.
voltage
Output V.(b)
(b)Inductor current iiLL,,A.
Inductorcurrent A.

4.2. LoadOnResistance
the contrary, the proposed control method tracks the reference voltage in 10 ms
Variations
without overshoot. The behavior of inductor current as shown in Figure 5b also further
To verify the robustness of the above controllers toward load uncertainty, the ref-
illustrates the superiority of the proposed control method.
erence voltage remains unchanged at 12 V, and the load resistance is changed from 30Ω
Ω atResistance
to4.2.20Load t = 2 s . The load resistance variations are along the following settings:
Variations
To verify the robustness of the above controllers toward load uncertainty, the reference
30Ω
voltage remains unchanged at 12 V, and (=load
the t ∈ [0,is2)changed from 30 Ω to 20 Ω at
R0 ),resistance

R =  are along
t = 2s. The load resistance variations 2 the following settings: (58)
 20Ω(= R0 ), t ∈ [2,5]
30 Ω (=3R0 ), t ∈ [0, 2)

R= (58)
20 Ω (= 23 R0 ), t ∈ [2, 5]
The responses of vo and iL under such a variation are depicted in Figure 6a,b. It
can beThe
seenresponses of vo6a
from Figure and i L the
that under such a variation
proposed are depicted
control method in Figure
ensures 6a,b. It can
faster settling be
time,
seen from Figure 6a that the proposed control method ensures faster settling
with only a slight drop in output voltage waveform around the desired trajectory. The time, with
only reason
main a slightisdrop
thatinthe
output
DHLRNN voltage waveform
has around
accurately the desired
compensated the trajectory. The main
model uncertainty
owing to load variations, whereas the ABTSMC method yields about a 0.02 V owing
reason is that the DHLRNN has accurately compensated the model uncertainty voltageto
load variations, whereas the ABTSMC method yields about a 0.02 V voltage
drop due to the inability to eliminate the undesirable impact of the external disturb- drop due to
ances. Although the BSMC−RBFNN method produces a lower voltage drop during the
load during the occurrence of load variations, the RBFNN takes a longer time to tune
neural network parameters. As a result, it takes 50 ms to stabilize the output voltage and
there exists an adverse overshoot, which may damage the performance of the buck con-
Mathematics 2023, 11, 1094 17 of 21

the inability to eliminate the undesirable impact of the external disturbances. Although
the BSMC−RBFNN method produces a lower voltage drop during the load during the
Mathematics 2023,
Mathematics 11, 11,
2023, x FOR PEER
x FOR REVIEW
PEER REVIEW 24 20 of 24
20 of network
occurrence of load variations, the RBFNN takes a longer time to tune neural
parameters. As a result, it takes 50 ms to stabilize the output voltage and there exists an
adverse overshoot, which may damage the performance of the buck converter.

1.5ms
1.5ms 50ms
50ms
0.01V
0.01V 0.02V
0.02V 0.01V 0.74A
1ms 0.01V 1.5ms 0.74A
1ms 1.5ms

(a) (b)
(a) Figure 6. Response curves with load resistance variations. (a) Output (b)
voltage vo , V. (b) Inductor
current iL , A.
Figure6.6.Response
Figure Responsecurves
curveswith
withload
loadresistance
resistancevariations.
variations. (a)
(a) Output v
Output voltage voo, ,V.V.(b)
(b)Inductor
Inductor
current
currenti L ,iLA., A.
4.3. Reference Voltage Variations
The tracking
4.3. Reference Voltage performance
Variations of the above controllers is evaluated by changing the ref-
4.3. Reference
erence voltage Voltage
from 12 Variations
V to 15 V at t = 2 s , and the load resistance remains unchanged
The tracking performance of the above controllers is evaluated by changing the refer-
at 30 Ω . The reference voltage variations are along the following settings: by changing the ref-
ence voltage from 12 V to 15 V atoft the
The tracking performance = 2above
s, andcontrollers is evaluated
the load resistance remains unchanged at
30erence
Ω. The voltage
referencefrom 12 V to
voltage 15 V at are
variations t = 2along
s , andthethe load resistance
following settings:remains unchanged
at 30Ω . The reference voltage variations 12V, t ∈ [0, the
are along 2) following settings:
vref =  (59)
12 V, t ∈
15V, [2,5]
vre f = 
t∈ [0, 2)
(59)
15 V, t ∈t ∈
12V, [2,[0,
5] 2)
The performance of the above vcontrollers ref =  during the reference variations can be (59)
The performance of the above controllers 15V,during t ∈ [2,5]
the
found in Figure 7a,b. Figure 7a illustrates that the ABTSMC and the BSMC−RBFNN reference variations can be found
in methods
Figure 7a,b. Figureto7atrack
are found illustrates
the newthat the ABTSMC
reference trajectory andin the ms and−180
600 BSMC RBFNN methods are
ms, respec-
found The
tively. performance
toCompared
track the new of neural
the the above
with reference controllers
trajectory
network during
in 600method,
control ms andthe the
180 reference
ms,
ABTSMC variations
respectively.
method to can be
Compared
fails
withobtain
foundtheinsatisfactory
Figure
neural networkdynamic
7a,b. Figureperformance
control 7a and
thereveals
illustrates
method, that the
ABTSMC a slower
ABTSMC
method responseand
fails in the
to rejecting ref-
BSMC−RBFNN
obtain satisfactory
erence value
methods
dynamic are uncertainty,
found to
performance and while
track theanew
the
reveals proposed
slower controltrajectory
reference
response method successfully
in rejectingin 600 ms tracks
reference and theuncertainty,
180
value new
ms, respec-
reference
while
tively.the voltage
Compared in
proposed with 10 ms
control without
themethod an obvious
successfully
neural network voltage
tracks
control rise. Moreover,
the new
method, the corresponding
thereference
ABTSMC voltage
method in 10 msto
fails
inductor
without ancurrent
obvious as voltage
plotted in Figure
rise. 7b also achieves
Moreover, a faster response
the corresponding and smaller
inductor current peak
as plotted
obtain satisfactory dynamic performance and reveals a slower response in rejecting ref-
current under the proposed method. Hence, because of the combination of the DHL-
inerence
Figurevalue
7b also achieves a faster response and smaller peak current
uncertainty, while the proposed control method successfully tracks the new under the proposed
RNN and adaptive algorithm, the proposed control method has superior tracking per-
method.
reference Hence,
voltage because
in 10 msof the combination
without an obvious of the DHLRNN
voltage and adaptive
rise. Moreover, thealgorithm,
correspondingthe
formance.
proposed
inductor control
currentmethod
as plottedhasinsuperior
Figure 7b tracking performance.
also achieves a faster response and smaller peak
current under the proposed method. Hence, because of the combination of the DHL-
RNN and 0.6Vadaptive algorithm, the proposed control method has superior tracking per-
formance. 1.2V

30ms 4A 4A 4.3A

0.6V
180ms
600ms 40ms 10ms
1.2V 20ms

30ms 4A 4A 4.3A

180ms
600ms 40ms 10ms
(a) 20ms (b)

Figure
Figure 7. 7.Response
Response curves
curves with
withreference
referencevoltage variations.
voltage (a) Output
variations. vo , V. (b)
voltagevoltage
(a) Output vo , Inductor
V. (b) Inductor
current i
current i L , A.
L , A.

(a) (b)

Figure 7. Response curves with reference voltage variations. (a) Output voltage vo , V. (b) Inductor
 fluences, the input voltage is difficult to maintain at the nominal value in the practical
converter system, and the input voltage will fluctuate around the nominal value. In or-
der to further investigate the steady−state performance of the above controllers toward
Mathematics 2023, 11, x FOR PEER REVIEW 21 of 24
the input voltage variations, a sawtooth disturbance with a period of 300 ms and ampli-
Mathematics 2023, 11, 1094 tude of 2 V is introduced to simulate the input voltage fluctuations. The reference volt- 18 of 21
age remains unchanged at 12 V, and the load resistance remains unchanged at 30Ω .
4.4. Input
Figure Voltage
8 shows theVariations
actual input voltage waveform in the presence of sawtooth disturb-
ance. Due to the long−term operation of the power source and external disturbance in-
4.4. Input Voltage Variations
Figure the
fluences, 9a,binput
showvoltage
the responses of vto
is difficult o and iL . When
maintain at thedealing
nominal with the time−varying
value in the practical
Due to
converter the long
system, and −the
terminput
operation
voltage ofwill
the fluctuate
power sourcearound andtheexternal
nominal disturbance
value.fluc- in-
In or-
input voltage fluctuations, the BSMC−RBFNN method produces obvious sawtooth
fluences, the
der to However, input voltage
further investigate is difficult
the voltage
steady−stateto maintain
performance at the nominal value in the practical
tuation. the output waveforms of the ofABTSMC
the above controllers
and toward
the proposed
converter
the system,variations,
input voltage and the input voltage will
a sawtooth fluctuatewith
disturbance around the nominal
a period of 300 value.
ms andInampli-
order
ABTSMC−DHLRNN methods are almost unchanged. The main reason is that both con-
to further
tude of investigate
2 V is introducedthe steady −
to simulate state performance of the above controllers toward the
trollers adopt parameter adaptive laws the input voltage
to estimate fluctuations.
the control The reference
coefficient, volt-
and a stable
input voltage variations, a sawtooth disturbance with a period of 300 ms and amplitude of
ageaccurate
and remainscontrol
unchanged
law is at 12 V, and
obtained the load
to remove theresistance
time−varyingremains
inputunchanged at 30
effects. It can beΩ .
2 V
Figure is introduced to simulate the input voltage fluctuations. The reference voltage remains
seen that8theshows the actual
proposed input
control voltagepossesses
method waveformbetterin thesteady−state
presence of tracking
sawtoothperfor-
disturb-
unchanged at 12 V, and the load resistance remains unchanged at 30 Ω. Figure 8 shows the
ance.and effective compensation for the input disturbances.
mance
actual input voltage waveform in the presence of sawtooth disturbance.
Figure 9a,b show the responses of vo and iL . When dealing with the time−varying
input voltage fluctuations, the BSMC−RBFNN method produces obvious sawtooth fluc-
tuation. However,300ms the output voltage waveforms of the ABTSMC and the proposed
ABTSMC−DHLRNN methods are almost unchanged. The main reason is that both con-
trollers adopt parameter adaptive laws to estimate the control coefficient, and a stable
and accurate control law is obtained to remove the time−varying input effects. It can be
seen that the 4V proposed control method possesses better steady−state tracking perfor-
mance and effective compensation for the input disturbances.

300ms

Figure 8. Waveform
Figure of of
8. Waveform real input
real input voltageVinVin
voltage with triangle
with disturbance.
triangle disturbance.
4V
Figure 9a,b show the responses of vo and i L . When dealing with the time−varying
input voltage fluctuations, the BSMC−RBFNN method produces obvious sawtooth fluc-
12.004 tuation. However, the output voltage waveforms of the ABTSMC and the proposed
12
ABTSMC−DHLRNN methods are almost unchanged. The main reason is that both con-
trollers adopt parameter adaptive laws to estimate the control coefficient, 0.12Aand a stable and
11.996
accurate
1.1 1.3
control
1.5
law is obtained to remove the time − varying input effects. It can be seen
that the proposed control method possesses better steady − state
Figure 8. Waveform of real input voltage Vin with triangle disturbance.tracking performance and
effective compensation for the input disturbances.
2mV 2mV
30ms 10mV 0.16A 0.16A

12.004

12

(a) (b) 0.12A

11.996
1.1 1.3 1.5
Figure 9. Response curves with input voltage variations. (a) Output voltage vo , V. (b) Inductor
current iL , A.
2mV 2mV
30ms 10mV 0.16A 0.16A

4.5. Comparison Analysis and Summary

In order to analyze and compare the performance of different controllers qualita-
tively, the performance indices are very useful, including maximum voltage
(a) (b) They were widely used
rise/maximum voltage drop (MVR/MVD) and settling time (ST).
Figure
Figure 9. Response
Response curves
curveswith
withinput
input voltage
voltagevariations.
variations.(a)
(a)Output voltagevvo o, , V.
Outputvoltage V. (b)
(b) Inductor
Inductor
current
current iiL
L
, A.
, A.
4.5. Comparison Analysis and Summary
4.5. Comparison Analysis and Summary
In order to analyze and compare the performance of different controllers qualitatively,
In order to analyze
the performance indices and
are compare the including
very useful, performance of different
maximum controllers
voltage qualita-
rise/maximum
tively, the performance indices are very useful, including maximum
voltage drop (MVR/MVD) and settling time (ST). They were widely used for comparison voltage
rise/maximum voltageTable
purposes in [44,45]. drop3(MVR/MVD) and settling
lists the comparison timeof
results (ST). They wereindices.
performance widely used
It is
observed in Table 3 that the proposed control method ensures optimal indices in tracking
Mathematics 2023, 11, 1094 19 of 21

the reference voltage. Therefore, it can be confirmed that the proposed control method has
better tracking accuracy and faster transient response speed, which improves the dynamic
and steady−state performance of the converter system.

Table 3. Performance indices comparison in above situations.

Performance Indices
Test Controllers
MVR/MVD (V) ST (ms)
ABTSMC −/1.2 −/260
1 BSMC−RBFNN 0.8/− 130/−
ABTSMC−DHLRNN 0/− 10/−
ABTSMC −/0.02 −/1.5
2 BSMC−RBFNN −/0.01 −/50
ABTSMC−DHLRNN −/0.01 −/1
ABTSMC −/1.2 −/600
3 BSMC−RBFNN 0.6/− 180/−
ABTSMC−DHLRNN 0.06/− 20/−
ABTSMC 0.001/0.001 −/−
4 BSMC−RBFNN 0.006/0.004 −/−
ABTSMC−DHLRNN 0.001/0.001 −/−

5. Conclusions
In this paper, an ABTSMC−DHLRNN method is designed for a class of SISO nonlinear
system. To counteract and compensate for the system unknown uncertainty, the DHLRNN
and parameter adaptive laws are utilized to estimate the nonlinear function and control
coefficient, respectively. Based on the adaptive control theory, Lyapunov theorem, and
finite−time convergence, the proposed ABTSMC−DHLRNN algorithm is designed. Al-
though the nonlinear system operates under the presence of uncertainty and disturbances,
the higher tracking accuracy and faster dynamic responses can be ensured. Furthermore,
the proposed control method can guarantee the finite time convergence of the closed−loop
system rigorously. The DHLRNN combines the merits of FNN, RNN, and MLNN, which
obtains fewer neurons training and stronger approximation capacity. All the parameters of
the network can adjust to optimal values through the adaptive learning algorithm, and the
additional parameter adaptive laws adopt the expression form of the projection algorithm
to ensure the boundedness of control coefficient estimates to obtain a stable control signal.
Moreover, on the one hand, the backstepping design simplifies the control design steps;
on the other hand, TSMC is introduced to guarantee that the tracking error can converge
to zero in finite time. The proposed control method is applied to a second−order DC–DC
buck converter system. The simulation result fully demonstrates its effectiveness and
superiority in tracking the reference voltage. Considering other disturbed power systems,
the proposed control method is designed for universal nonlinear systems, which can be
directly applied without major modifications. The chattering problem still exists in SMC.
Therefore, in future work, we will concentrate on how to further deal with the chattering
problem of the control signal by using more advanced algorithms.

Author Contributions: Conceptualization, J.F.; methodology, X.G., W.F., and X.B.; writing—original
draft preparation, X.G.; writing—review and editing, J.F. All authors have read and agreed to the
published version of the manuscript.
Funding: This work is partially supported by the National Science Foundation of China under Grant
No. 62273131.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.
Mathematics 2023, 11, 1094 20 of 21

Nomenclature

PID Proportional−Integral−Derivative
SMC Sliding mode control
TSMC Terminal sliding mode control
BSMC Backstepping sliding mode control
ABTSMC Adaptive backstepping terminal sliding mode control
RBFNN Radial basis function neural network
FNN Fuzzy neural network
RNN Recurrent neural network
DHLRNN Double hidden recurrent neural network

References
1. Pan, Z.; Dong, F.; Zhao, J.; Wang, L.; Wang, H.; Feng, Y. Combined Resonant Controller and Two-Degree-of-Freedom PID
Controller for PMSLM Current Harmonics Suppression. IEEE Trans. Ind. Electron. 2018, 65, 7558–7568. [CrossRef]
2. Hakan, U.; Irfan, O.; Ugur, Y.; Metin, K. Test Platform and Graphical User Interface Design for Vertical Take-Off and Landing
Drones. Rom. J. Inf. Sci. Technol. (ROMJIST) 2022, 25, 350–367.
3. Lee, J.Y.; Jin, M.; Chang, P.H. Variable PID Gain Tuning Method Using Backstepping Control with Time-Delay Estimation and
Nonlinear Damping. IEEE Trans. Ind. Electron. 2014, 61, 6975–6985. [CrossRef]
4. Meza, J.L.; Santibanez, V.; Soto, R.; Llama, M.A. Fuzzy Self-Tuning PID Semiglobal Regulator for Robot Manipulators. IEEE Trans.
Ind. Electron. 2012, 59, 2709–2717. [CrossRef]
5. Precup, R.-E.; Preitl, S.; Rudas, I.J.; Tomescu, M.L.; Tar, J.K. Design and experiments for a class of fuzzy controlled servo systems.
IEEE/ASME Trans. Mechatron. 2008, 13, 22–35. [CrossRef]
6. Incremona, G.P.; Rubagotti, M.; Ferrara, A. Sliding Mode Control of Constrained Nonlinear Systems. IEEE Trans. Autom. Control
2017, 62, 2965–2972. [CrossRef]
7. Zheng, X.; Yang, X. Command Filter and Universal Approximator Based Backstepping Control Design for Strict-Feedback
Nonlinear Systems with Uncertainty. IEEE Trans. Autom. Control 2020, 65, 1310–1317. [CrossRef]
8. Cai, J.; Wen, C.; Su, H.; Liu, Z.; Xing, L. Adaptive Backstepping Control for a Class of Nonlinear Systems with Non-Triangular
Structural Uncertainties. IEEE Trans. Autom. Control 2017, 62, 5220–5226. [CrossRef]
9. Lin, F.-J.; Huang, M.-S.; Chen, S.-G.; Hsu, C.-W.; Liang, C.-H. Adaptive Backstepping Control for Synchronous Reluctance Motor
Based on Intelligent Current Angle Control. IEEE Trans. Power Electron. 2020, 35, 7465–7479. [CrossRef]
10. Ferrara, A.; Incremona, G.P. Design of an Integral Suboptimal Second-Order Sliding Mode Controller for the Robust Motion
Control of Robot Manipulators. IEEE Trans. Control. Syst. Technol. 2015, 23, 2316–2325. [CrossRef]
11. Fei, J.; Feng, Z. Fractional-Order Finite-Time Super-Twisting Sliding Mode Control of Micro Gyroscope Based on Double-Loop
Fuzzy Neural Network. IEEE Trans. Syst. Man, Cybern. Syst. 2021, 51, 7692–7706. [CrossRef]
12. Huang, S.; Wang, J.; Xiong, L.; Liu, J.; Li, P.; Wang, Z.; Yao, G. Fixed-Time Backstepping Fractional-Order Sliding Mode Excitation
Control for Performance Improvement of Power System. IEEE Trans. Circuits Syst. I Regul. Pap. 2022, 69, 956–969. [CrossRef]
13. Zhang, J.; Sun, C.; Zhang, R.; Qian, C. Adaptive sliding mode control for re-entry attitude of near space hypersonic vehicle based
on backstepping design. IEEE/CAA J. Autom. Sin. 2015, 2, 94–101.
14. Chen, F.; Jiang, R.; Zhang, K.; Jiang, B.; Tao, G. Robust Backstepping Sliding Mode Control and Observer-Based Fault Estimation
for a Quadrotor UAV. IEEE Trans. Ind. Electron. 2016, 63, 5044–5056. [CrossRef]
15. Alipour, M.; Zarei, J.; Razavi-Far, R.; Saif, M.; Mijatovic, N.; Dragicevic, T. Observer-Based Backstepping Sliding Mode Control
Design for Microgrids Feeding a Constant Power Load. IEEE Trans. Ind. Electron. 2023, 70, 465–473. [CrossRef]
16. Yu, X.; Man, Z. Fast terminal sliding-mode control design for nonlinear dynamical systems. IEEE Tran. Circuits Syst. I Fundam.
Theory Appl. 2002, 49, 261–264.
17. Yao, X.; Park, J.H.; Dong, H.; Guo, L.; Lin, X. Robust Adaptive Nonsingular Terminal Sliding Mode Control for Automatic Train
Operation. IEEE Trans. Syst. Man Cybern. Syst. 2019, 49, 2406–2415. [CrossRef]
18. Wang, H.; Zhang, Q.; Sun, Z.; Tang, X.; Chen, I.-M. Continuous Terminal Sliding-Mode Control for FJR Subject to
Matched/Mismatched Disturbances. IEEE Trans. Cybern. 2022, 52, 10479–10489. [CrossRef]
19. Xu, B.; Zhang, L.; Ji, W. Improved Non-Singular Fast Terminal Sliding Mode Control with Disturbance Observer for PMSM
Drives. IEEE Trans. Transp. Electrif. 2021, 7, 2753–2762. [CrossRef]
20. Oucheriah, S.; Guo, L. PWM-Based Adaptive Sliding-Mode Control for Boost DC–DC Converters. IEEE Trans. Ind. Electron. 2013,
60, 3291–3294. [CrossRef]
21. Chen, L.; Liu, M.; Huang, X.; Fu, S.; Qiu, J. Adaptive Fuzzy Sliding Mode Control for Network-Based Nonlinear Systems with
Actuator Failures. IEEE Trans. Fuzzy Syst. 2018, 26, 1311–1323. [CrossRef]
22. Xu, L.; Yao, B. Adaptive robust control of mechanical systems with non-linear dynamic friction compensation. Int. J. Control.
2008, 81, 167–176. [CrossRef]
23. Zamfirache, I.A.; Precup, R.-E.; Roman, R.-C.; Petriu, E.M. Reinforcement Learning-based control using Q-learning and gravita-
tional search algorithm with experimental validation on a nonlinear servo system. Inf. Sci. 2022, 583, 99–120. [CrossRef]
Mathematics 2023, 11, 1094 21 of 21

24. Yahui, L.; Sheng, Q.; Xianyi, Z.; Kaynak, O. Robust and adaptive backstepping control for nonlinear systems using RBF neural
networks. IEEE Tran. Neural Netw. 2004, 15, 693–701.
25. Lian, J.; Lee, Y.; Sudhoff, S.D.; Zak, S.H. Self-Organizing Radial Basis Function Network for Real-Time Approximation of
Continuous-Time Dynamical Systems. IEEE Trans. Neural Netw. 2008, 19, 460–474. [CrossRef]
26. Fei, J.; Chen, Y.; Liu, L.; Fang, Y. Fuzzy Multiple Hidden Layer Recurrent Neural Control of Nonlinear System Using Terminal
Sliding-Mode Controller. IEEE Trans. Cybern. 2022, 52, 9519–9534. [CrossRef]
27. Fei, J.; Wang, H.; Fang, Y. Novel Neural Network Fractional-order Sliding-Mode Control with Application to Active Power Filter.
IEEE Trans. Syst. Man Cybern. Syst. 2022, 52, 3508–3518. [CrossRef]
28. El-Sousy, F.; Amin, M.; Mohammed, O. Robust Optimal Control of High-Speed Permanent-Magnet Synchronous Motor Drives
via Self-Constructing Fuzzy Wavelet Neural Network. IEEE Tran. Ind. Appl. 2021, 57, 999–1013. [CrossRef]
29. Fei, J.; Wang, Z.; Fang, Y. Self-Evolving Chebyshev Fuzzy Neural Fractional-Order Sliding Mode Control for Active Power Filter.
IEEE Trans. Ind. Inform. 2022. [CrossRef]
30. Lin, F.-J.; Hung, Y.-C.; Tsai, M.-T. Fault-Tolerant Control for Six-Phase PMSM Drive System via Intelligent Complementary
Sliding-Mode Control Using TSKFNN-AMF. IEEE Trans. Ind. Electron. 2013, 60, 5747–5762. [CrossRef]
31. Han, H.-G.; Zhang, L.; Hou, Y.; Qiao, J.-F. Nonlinear Model Predictive Control Based on a Self-Organizing Recurrent Neural
Network. IEEE Trans. Neural Netw. Learn. Syst. 2016, 27, 402–415. [CrossRef]
32. Fei, J.; Lu, C. Adaptive Sliding Mode Control of Dynamic Systems Using Double Loop Recurrent Neural Network Structure.
IEEE Trans. Neural Netw. Learn. Syst. 2018, 29, 1275–1286. [CrossRef]
33. El-Sousy, F.; Abuhasel, K. Adaptive Nonlinear Disturbance Observer Using a Double-Loop Self-Organizing Recurrent Wavelet
Neural Network for a Two-Axis Motion Control System. IEEE Tran. Ind. Appl. 2018, 54, 764–786. [CrossRef]
34. Lee, H.; Jung, M.; Tani, J. Recognition of Visually Perceived Compositional Human Actions by Multiple Spatio-Temporal Scales
Recurrent Neural Networks. IEEE Trans. Cogn. Dev. Syst. 2018, 10, 1058–1069. [CrossRef]
35. Chen, H.; Wu, C.; Du, B.; Zhang, L.; Wang, L. Change Detection in Multisource VHR Images via Deep Siamese Convolutional
Multiple-Layers Recurrent Neural Network. IEEE Trans. Geosci. Remote Sens. 2020, 58, 2848–2864. [CrossRef]
36. Makondo, W.; Nallanthighal, R.; Mapanga, I.; Kadebu, P. Exploratory test oracle using multi-layer perceptron neural network.
In Proceedings of the 2016 International Conference on Advances in Computing, Communications and Informatics (ICACCI),
Jaipur, India, 21–24 September 2016; pp. 1166–1171.
37. Xu, B.; Zhang, R.; Li, S.; He, W.; Shi, Z. Composite Neural Learning-Based Nonsingular Terminal Sliding Mode Control of MEMS
Gyroscopes. IEEE Trans. Neural Netw. Learn. Syst. 2020, 31, 1375–1386. [CrossRef]
38. Fei, J.; Chen, Y. Fuzzy Double Hidden Layer Recurrent Neural Terminal Sliding Mode Control of Single-Phase Active Power
Filter. IEEE Trans. Fuzzy Syst. 2021, 29, 3067–3081. [CrossRef]
39. Li, Y.; Zhang, J.; Liu, W.; Tong, S. Observer-based adaptive optimized control for stochastic nonlinear systems with input and
state constraints. IEEE Trans. Neural Netw. Learn. Syst. 2021, 33, 7791–7805. [CrossRef]
40. Wang, Z.; Fei, J. Fractional-Order Terminal Sliding-Mode Control Using Self-Evolving Recurrent Chebyshev Fuzzy Neural
Network for MEMS Gyroscope. IEEE Trans. Fuzzy Syst. 2022, 30, 2747–2758. [CrossRef]
41. Hua, M.; Zheng, D.; Deng, F. H∞ filtering for nonhomogeneous Markovian jump repeated scalar nonlinear systems with
multiplicative noises and partially mode-dependent characterization. IEEE Trans. Syst. Man Cybern. Syst. 2021, 51, 3180–3192.
[CrossRef]
42. Fei, J.; Chen, Y. Dynamic Terminal Sliding-Mode Control for Single-Phase Active Power Filter Using New Feedback Recurrent
Neural Network. IEEE Trans. Power Electron. 2020, 35, 9904–9922. [CrossRef]
43. Fei, J.; Wang, Z.; Pan, Q. Self-Constructing Fuzzy Neural Fractional-Order Sliding Mode Control of Active Power Filter. IEEE
Trans. Neural Netw. Learn. Syst. 2022, 1–12. [CrossRef]
44. Babes, B.; Mekhilef, S.; Boutaghane, A.; Rahmani, L. Fuzzy Approximation-Based Fractional-Order Nonsingular Terminal Sliding
Mode Controller for DC–DC Buck Converters. IEEE Trans. Power Electron. 2022, 37, 2749–2760. [CrossRef]
45. Lin, X.; Liu, J.; Liu, F.; Liu, Z.; Gao, Y.; Sun, G. Fractional-Order Sliding Mode Approach of Buck Converters with Mismatched
Disturbances. IEEE Trans. Circuits Syst. I Regul. Pap. 2021, 68, 3890–3900. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.