Original Paper

Measurement and Control

2021, Vol. 54(7-8) 1245–1258
Observer-based adaptive neural Ó The Author(s) 2021
Article reuse guidelines:
network backstepping sliding mode sagepub.com/journals-permissions
DOI: 10.1177/00202940211021107

control for switched fractional order journals.sagepub.com/home/mac

uncertain nonlinear systems with

unmeasured states

Tao Chen , Damin Cao, Jiaxin Yuan and Hui Yang

Abstract
This paper proposes an observer-based adaptive neural network backstepping sliding mode controller to ensure the sta-
bility of switched fractional order strict-feedback nonlinear systems in the presence of arbitrary switchings and unmea-
sured states. To avoid ‘‘explosion of complexity’’ and obtain fractional derivatives for virtual control functions
continuously, the fractional order dynamic surface control (DSC) technology is introduced into the controller. An obser-
ver is used for states estimation of the fractional order systems. The sliding mode control technology is introduced to
enhance robustness. The unknown nonlinear functions and uncertain disturbances are approximated by the radial basis
function neural networks (RBFNNs). The stability of system is ensured by the constructed Lyapunov functions. The frac-
tional adaptive laws are proposed to update uncertain parameters. The proposed controller can ensure convergence of
the tracking error and all the states remain bounded in the closed-loop systems. Lastly, the feasibility of the proposed
control method is proved by giving two examples.

Keywords
Fractional order nonlinear systems, adaptive backstepping control, dynamic surface control, switched systems, neural
network, observer

Date received: 14 January 2021; accepted: 29 April 2021

Introduction unknown external disturbances and input uncertainties.

Jain et al.9 proposed a fractional order internal model
Fractional calculus has unique memory properties and control method for non-ideal dc–dc buck and boost
the ability to accurately model the system, so it can be converter.
easily applied to many industries and research fields, Adaptive backstepping control (ABC) method has
such as research electrical circuits, signal processing, been widely used to stabilize nonlinear systems in prac-
chemical processes and biological engineering.1–4 tical applications due to the excellent performance of
Fractional order dynamic system can reflect the sys- compensating saturation.10 For example, Zhou et al.10
tem’s own conditions more truly, so as to develop new studied the stabilization problem for uncertain non-
control strategies to enhance the characteristics of con- linear systems and proposed an ABC algorithm. Sun
trol loops.5 In recent years, many important results had et al.11 proposed an ABC method to handle the posi-
been achieved in the stability analysis and control tion tracking problem of the permanent magnet syn-
research of fractional order nonlinear systems. For chronous motor system. Cai et al.12 proposed a new
example, Wang6 extended backstepping control scheme ABC scheme to overcome the uncertainties for a class
to fractional order system and studied problem of
Mittage-Leffler stabilization of fractional order non-
linear system. Ding et al.7 proposed a fractional order School of Air Transportation, Shanghai University of Engineering Science,
backstepping controller for a class of fractional order Shanghai, China
nonlinear strict-feedback system with both unknown
Corresponding author:
disturbance. Li et al.8 designed an adaptive sliding Damin Cao, School of Air Transportation, Shanghai University of
mode controller to compensate the input uncertainties Engineering Science, Shanghai 201620, China.
for a class of fractional order nonlinear systems with Email: [email protected]

Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License
(https://creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
1246 Measurement and Control 54(7-8)

of second-order nonlinear systems with non-triangular PMSM with the immeasurable state and the parameter
uncertainties. At present, some control methods are uncertainties. Tong et al.29 proposed a fuzzy state
combined with adaptive backstepping control methods observer to estimate the unmeasurable states and
to study classical nonlinear systems in the field of designed an adaptive fuzzy output-feedback backstep-
integer-order control method, such as adaptive back- ping controller for uncertain strict-feedback nonlinear
stepping sliding mode control (SMC),13 adaptive neural systems. For fractional order systems with system
network, or fuzzy backstepping,14,15 adaptive dynamic uncertainties and external disturbances, observers and
surface backstepping control,16 and so on. NN approximations are rarely considered together.
Traditional backstepping control requires accurate This is of great significance for further research on new
modelling information of the controlled object and can- control technologies based on neural networks and
not overcome the disturbance. To deal with the uncer- observers.
tain of nonlinear systems, the universal approximation It should be recognized that the abovementioned
theories of neural networks (NNs) and fuzzy logic sys- fractional order nonlinear systems are a class of non-
tems (FLSs) can be employed to approach the unknown switched systems. It is worth pointing out that the
nonlinear functions.17 For example, the authors pro- switched system is another more complicated system. It
posed a fuzzy double hidden layer recurrent neural net- is formed by switching signals between subsystems.
work approximation technology in Refs.,18–20 which Practical applications, such as hybrid vehicles and cir-
can be regarded as a combination of a fuzzy NN and a cuit systems, are all switching systems. Applying the
RBFNN to improve the accuracy of a nonlinear common Lyapunov function method, Tong et al.30
approximation, so it has the advantages of these two solved the control problem of arbitrary switching sys-
NNs. Aiming at the field of fractional order systems tems. Sui et al.,27 Long and Zhao31 and Li and Tong32
control, Liu et al.21 designed adaptive fuzzy backstep- studied the switching control approaches of strict feed-
ping control method is proposed for uncertain frac- back switched nonlinear systems through the average
tional order chaotic systems including unknown dwell time method.
external disturbance and input saturation. Wang and Based on the previous discussion, this paper pro-
Liang22 combined backstepping and adaptive technol- poses an adaptive neural network backstepping sliding
ogy, using RBFNN to approximate the system mode controller based on DSC technology for a class
unknown nonlinear uncertainty and proposed an ABC of the switched fractional order strict-feedback non-
method for a class of uncertain fractional nonlinear sys- linear systems with unmeasured state. It should be
tems with external disturbance and input saturation. pointed out that the theoretical results obtained in this
Ma and Ma23 proposed to introduce an auxiliary func- paper is not a simple extension from integer-order sys-
tion to compensate the unknown external interference tems to fractional systems. We use some properties of
and the approximation error produced by the FLS the Caputo fractional derivative and the integral
approximation of the unknown function. Zhang and inequality to overcome the adverse effects from the
Li13 proposed an adaptive backstepping sliding mode incorporation of weakly singular kernels in fractional
controller based on NN technology for the wheel slip derivative. Compared with the current researches, the
tracking control system. For fractional multi-agent sys- contributions of this work are list as follows:
tems with unknown uncertainty, Shahvali et al.24
designed a distributed controller based on NN control 1) In comparison with Refs.,21,22,28 this paper pro-
method, the NN is used to approximate the system poses an observer-based adaptive neural net-
unknown uncertainty. Based on RBFNN, an adaptive work backstepping sliding mode controller for a
backstepping control method was proposed by class of the switched fractional order strict-
Shahvali et al.25 for the consensus problem of fractional feedback nonlinear systems with unmeasured
order nonlinear multi-agent systems. For fractional states. Compared with the previous works in
order input saturated permanent magnet synchronous Sui et al.,27 the state observer is introduced into
motor (PMSM), Lu and Wang26 proposed a NN the proposed method to estimate the system
method combined with command filtering technique. state.
Sui et al.27 used NN to model uncertain fractional order 2) Compared with the previous works in Ma and
systems and presented an adaptive switching dynamic Ma,23 the sliding mode control term is intro-
surface control method for fractional order non-strict duced to enhance robustness and the fractional
feedback nonlinear system. order DSC technology is used to avoid ‘‘explo-
In most practical applications, the system states can- sion of complexity’’ and obtain fractional deri-
not be fully accessed, and the designer cannot know the vatives for virtual control continuously.
output and input of the plant. In this case, observer- Compared with the previous works in Ding
based control is usually required. There have been et al.,7 the uncertain disturbances are approxi-
many studies in the field of integer-order control com- mated by RBFNN. The stability of the closed-
bining observers with adaptive inversion techniques. loop system is ensured by the constructed
For example, Lu and Wang28 designed a state observer Lyapunov functions. And in our design, the
to obtain the unmeasured state for a fractional chaotic proposed fractional update laws estimate the
Chen et al. 1247

unknown parameters and the upper limit of the   sag

L tg1 Ea, g ð1Þ = a ð5Þ
approximation errors. s +a
3) Under the framework of adaptive backstepping Lemma 1. Podlubny34 For a complex number b and
control technique, the desired tracking perfor- two real numbers a, v satisfying a 2 ð0, 1Þ and
mance for the switched fractional order strict-
feedback nonlinear systems is obtained by using pa
\ v \ minfp, pag ð6Þ
the designed controller which integrates obser- 2
ver, RBFNN, fractional order adaptive laws, For all integer n51, we can obtain
sliding mode control method, DSC technique. !
X
‘
1 1
The rest of the paper is organized as follows. Section Ea, b ð1Þ = +o n+1
ð7Þ
j=1
G ð b  aj Þ j 1j
2 introduces basic theory about fractional calculus and
proposes the controlled switched fractional order non- when j1j ! ‘, v4jargð1Þj4p.
linear system. In Section 3, we construct an observer to
estimate the system states firstly, and then a controller Lemma 2. Podlubny34 Let a 2 ð0, 2Þ and b be an arbi-
is proposed based on the adaptive backstepping sliding trary real number, and for 8y . 0 such that
mode control method. In Section 4, the effectiveness of ðpa=2Þ \ y4 minfp, pag, one has
the proposed control method is proved by giving two  
Ea, b ð1Þ4 m
examples. In Section 5, we summarize the paper and ð8Þ
1 + j 1j
give some conclusions.
where m . 0, y4jargð1Þj4p, and j1j50.
Preliminaries Lemma 3. Duarte-Mermoud et al.35 Let xðtÞ = ½x1
Fractional calculus ðtÞ, :::, xn ðtÞ 2 Rn be a vector of continuous and differ-
entiable function. And then, the following relationship
The Riemann-Liouville (R-L) fractional derivative defi-
holds
nition and Caputo fractional derivative definition are
commonly used in the study of fractional differential. 1 a T 
Because Caputo fractional derivative definition is easier D x ðtÞPx 4xT ðtÞPDa xðtÞ
2 ð9Þ
to give the initial value conditions of the fractional deri- 8a 2 ð0, 1Þ, 8t . t0
vative equation, it is widely used in engineering applica-
tions. This paper mainly adopts the Caputo definition. Lemma 4. Wang et al.36 For any x, y 2 Rn , the follow-
The R-L fractional integral5 is defined as ing inequality relationship holds
ð
1 t fðt Þ ca 1
RL a
0 It fðtÞ = dt ð1Þ xT y4 kxka + b kykb ð10Þ
GðaÞ 0 ðt  t Þ1a a bc
where a . 1, b . 1, c . 0, and ða  1Þðb  1Þ = 1.
The R-L fractional derivative is defined as
ð Lemma 5. Li et al.37 In fractional-order nonlinear sys-
RL a 1 dn t fðt Þ
0 It f ðt Þ = dt ð2Þ tem, if the a-order derivative of Lyapunov function
Gðn  aÞ dtn 0 ðt  t Þ1 + an
Vðt, xÞ satisfying
Б
where n 2 N and n  1 \ a4n, GðzÞ = 0 tz1 et dt is Da Vðt, xÞ4  CVðt, xÞ + D
the Gamma function.
The Caputo fractional derivative5 is defined as we can obtain
ðt Dm
C a 1 fðnÞ ðt Þ Vðt, xÞ4Vð0ÞEa ðCta Þ + , t50 ð11Þ
0 Dt fðtÞ = dt ð3Þ C
Gðn  aÞ 0 ðt  t Þ1 + an
where 0 \ a \ 1, C . 0, and D50. Then, Vðt, xÞ is
Remark 1. To simplify the notation, we set
C a a bounded on ½0, t and fractional order systems are sta-
0 Dt fðtÞ = D fðtÞ. ble, where m is defined in Lemma 2
Definition 1. Li et al.33 The Mittag–Leffler function is
given as System description
X
‘
1 k Many fractional order nonlinear systems can be written
Ea, g ð1Þ = ð4Þ
Gðak + gÞ as a lower triangular form, and it is referred as the
k=0
lower triangular system. For example, the fractional
where 1 is complex number and a, g . 0. Its Laplace order Chua-Hartley’s system Wang et al.,38 the frac-
transform is given by tional order Arneodo system39 and the fractional order
1248 Measurement and Control 54(7-8)

Duffing-Holmes chaotic system.40 The fractional order controller based on observer, and choose suitable para-
Arneodo system: meters adaptive laws such that all the signals remain
8 a bounded and the tracking error S1 is as small as possi-
< D xðtÞ = yðtÞ ble in the closed-loop system.
Da yðtÞ = zðtÞ ð12Þ
: a
D zðtÞ = axðtÞ  byðtÞ  rzðtÞ  x3 ðtÞ
Main results
the fractional order Chua-Hartley’s system:
8  
Observer design
< Da x1 ðtÞ = x2 ðtÞ + a x1 ðtÞ  x31 ðtÞ Assumption 1. The unknown functions fi ðXÞ, i = 1, :::, n
Da x ðtÞ = x3 + bx1 ðtÞ  cx2 ðtÞ ð13Þ
: a 2 can be expressed as
D x3 ðtÞ = dx2 ðtÞ
fi ðXi jui Þ = uTi ui ðXi Þ, 14i4n ð18Þ
the fractional order Duffing-Holmes system:
 a where ui is the ideal constant vector, ui ðXi Þ is the basis
D x1 = x2 functions vector, and Gaussian basis functions are used
ð14Þ
Da x2 = ax1  bx2  cx31 in this paper. In this section, assuming that the state
In the paper, we consider the switched fractional order variables of the system (15) are not available. In this
strict-feedback nonlinear system with uncertain case, the system states need to be estimated by an obser-
disturbances. ver, and the observer is designed as
8 P
n  
> a s ðt Þ s ðt Þ
Da X^ = AX^ + Ky + Bi fiq X^i jui + Bu
> D x1 ðtÞ = x2 + f1 sððx
>
< tÞ
1 Þ + d1 ðtÞ
s ð tÞ i=1 ð19Þ
Da xi ðtÞ = xi + 1 + fi ðx1 , x2 , :::, xi Þ + di ðtÞ ð15Þ
> a s ðt Þ s ðt Þ y^ = CX^
>
> D xn ðtÞ = u + fn ðx1 , x2 , :::, xn Þ + dn ðtÞ
:
y = x1 where X^i = ðx^1 , x^2 , :::, x^n ÞT are the estimated values of
Xi = ðx1 , x2 , :::, xn ÞT .
where i = 2, :::, n  1, a 2 ð0, 1Þ is the order of the frac- Let e = X  X^ be state observation errors of the sys-
tional derivative; Xi = ðx1 , x2 , :::, xi ÞT 2 Rn is the sys- tem, then according to equations (16) and (19), we have
tem states vector. uðtÞ is the control input of the
sðtÞ h   i
system, y is the system output and fi ðx1 , x2 , :::, xi Þ is X
n  
unknown nonlinear function; sðtÞ is a piecewise contin- Da e = Ae + Bi fqi X^i  f^qi X^i jui + Dfqi + dqi
i=1
uous function that is used to describe the triggering
conditions for switching between subsystems. It is ð20Þ
called a switching signal, for example, if sðtÞ = q, it  
where Dfqi = fqi ðXi Þ  fqi X^i .
means that q  th subsystem q is activated. dki ðtÞ is exter- By Assumption 1, we can obtain
 
nal disturbance and di ðtÞ 4di , where di is an

   
unknown constant. f^qi X^i jui = uTi ui X^i , 14i4n ð21Þ
Rewriting system (15):
Define the vectors of optimal parameters as
P
n
 
Da X = AX + Ky + Bi ½fqi ðXi Þ + dqi ðtÞ + Bu h
   i
i=1 u = arg min sup ^ f^q X^i jui  fq X^i 
i Xi 2Ui i i ð22Þ
y = CX ui 2Oi

ð16Þ where 14i4n, Oi and Ui are compact regions for ui , Xi

2 3 2 3 and X^i .
k11 k11 Define errors of the optimal approximation and
6 . 7 6 . 7
Where A = 4 .. I 5, , K = 4 .. 5, parameters estimation as
k1n 0  0 k1n     
 T
2 3T eqi = fqi X^i  f^qi X^i ui
ð23Þ
0...0 1 ~ui = u  ui , i = 1, 2, :::, n
B = |fflffl{zfflffl} , Bi = 40|fflffl{zfflffl}
. . . 1 . . . 05 , i
n1
 
i Assumption 2. The optimal approximation error and
1 0 ...0
|fflffl{zfflffl} the ideal constant weight vector remain bounded,
 there
C= , and given a positive matrix QT = Q,
n1 exists positive constants e0 , ui satisfying eqi 4e0 ,
there exists a positive matrix PT = P satisfying jui j4ui

AT P + PA = 2Q ð17Þ Assumption 3. There exists a set of known constants mi ,

the following relationship holds
Control objectives: Let yd be a reference signal and the
tracking error as S1 = y  yd , design fractional order jfi ðX1 Þ  fi ðX2 Þj4mi kX1  X2 k ð24Þ
Chen et al. 1249

By equations (20) and (23), we have Controller design

X
n h     i Theorem 1. For fractional order nonlinear system (15)
a
D e = Ae + Bi fqi X^i  f^qi X^i jui + Dfqi + dqi where Assumption 2 holds, the following designs can
i=1 ensure that the tracking error tends to a small region
X
n  
of the origin and the closed-loop system is semi-global
= Ae + Bi eqi + Dfqi + ~
uTi ui X^i + dqi stability,
i=1
the error surfaces
X
n T  

= Ae + Df + k + ~ ui X^i
Bi u 
i S 1 = y  yd
i=1 ð31Þ
Si = x^i  vi , i = 2, :::, n
ð25Þ

T
T the intermediate control functions
where k = eq1 + dq1 , :::, eqn + dqn , Df = Dfq1 , :::, Dfqn .
8

Constructed the first Lyapunov function: > b1 = 2k1 S1 + uT1 u1 ðx^1 Þ + signðS1 Þd1  D3
a
yd
>
<
ki Si + Si1 + uTi ui + signðSi Þdi
1 T 4 5
V0 = e Pe ð26Þ >
> b = b  vi
2 : i  i1
li
According to Lemma 3 and (17), we can obtain ð32Þ
1   the parameters update laws
Da V0 4 eT PAT + AP e + eT Pðk + DfÞ
2  a
Xn T  
D ui = si ui ðx^1 , x^1 , :::, x^i ÞSi  ri ui
ð33Þ
+ eT PBi ~
ui ui X^i Da di = ri jSi j  hi di
i=1 ð27Þ
T
4  e Qe + e Pðk + DfÞ T the control input
0 1
X
n   Sn1 + kn Sn + uTn un + signðSn Þdn
+ eT P Bi ~
uTi ui X^i
u = @ b  vn A ð34Þ
i=1  n1
ln
By Lemma 4 and Assumption 3, we can obtain
where ~ui = ui  ui is the parameter estimation error,
    ~di = d  di is the upper bound estimation error,
eT Pðk + DfÞ4eT Pk + eT PDf i
signðSi Þ is the sliding term to enhance robustness, si
1 1 1 1
4 kek2 + kPkk2 + kek2 + kPk2 kDfk2 and ri are design constant parameter, ki . 0, si . 0,
2 2 2 2 ri . 0, ri . 0, hi . 0.
1 1 Xn  
2  2
4kek + kPk k + kPk 2 Dfq 2
2 2 i Proof. Step 1. the first tracking error surface is defined
i=1 ð28Þ as
1 X n
2 1
4 ke k2 + kek2 kPk2 mqi + kPk k2 S1 = y  y d ð35Þ
2 i=1
2
!
1 X n
2 1 According to e2 = x2  x^2 , we can obtain x2 = x^2 + e2 ,
4 ke k2 1 + kPk2 mqi + kPk k2 and we have
2 i=1
2
Da S1 = Da x1  Da yd
and
= x2 + fq1 ðx1 Þ + dq1  Da yd
X
n   = x^2 + fq1 ðx1 Þ + dq1  Da yd + e2 ð36Þ
eT P Bi ~
uTi ui X^i
i=1 = x^2 + uT u1 ðx^1 Þ + ~uT u1 ðx^1 Þ
1 1
1X n      Da yd + dq1 + eq1 + Dfq1 + e2
4teT PT Pe + uT u X^i uTi X^i ~
~ ui ð29Þ
t i=1 i i
Define the second error surface S2 and the output error
1X n
~ 2 of a fractional order filter
4tl2max ðPÞkek + uT ~
ui
t i=1 i
S2 = x^2  v2

ð37Þ
where k = e10 + d1 , . . . , en0 + dn , t . 0.
 w2 = v2  b1
By equations (27)–(29), we obtain Then we obtain

1 1X n
Da S1 = S2 + w2 + b1 + uT1 u1 ðx^1 Þ + ~uT1 u1 ðx^1 Þ
Da V0 4  q0 kek2 + kPk k2 + ~
uT ~ui ð30Þ ð38Þ
2 t i=1 i + dq1 + eq1 + Dfq1 + e2  Da yd
!
1 X
n
1 2
where q0 = lmin ðQÞ  1 + kPk2 mq2
i + lmax ðPÞ . Construct the second Lyapunov function:
2 i=1
t
1250 Measurement and Control 54(7-8)

V1 = V0 +
1 2
S1 +
1 ~T ~
u1 u1 +
1 ~2
d ð39Þ 1 1X n

2 2s1 2r1 1 D a V 1 4  q0 ke k 2 + kPk k2 + ~uT ~ui

2 t i=1 i
From (30) and (38), we have  k1 S21 + S1 S2  jS1 jd1 + S1 w2 + S1 e2 + S1 d1
1 ~T a ~ 1 r h
Da V1 4Da V0 + S1 Da S1 + u D u+ ~ d1 Da ~d1 + 1 ~uT1 u1  jS1 j~d1 + 1 ~d1 d1
s1 1 r1 s1 r1
1X n 1 1X n
1
4  q0 kek2 + kPk k2 + ~
uT ~
ui 4  q0 kek2 + kPk k2 + ~uT ~ui
2 t i=1 i 2 t i=1 i
   k1 S21 + S1 S2  jS1 jd1 + S1 w2 + S1 e2
+ S1 S2 + w2 + b1 + uT1 uðx^1 Þ + ~ uT1 uðx^1 Þ  Da yd  
  1 T a + jS1 j ~d1 + d1
+ S1 e2 + S1 eq1 + Dfq1 + dq1  u ~ D u1  1 d~1 Da d1
s1 1 r1 r h
+ 1 ~uT1 u1  jS1 j~d1 + 1 ~d1 d1
2 1 2
s1 r1
4  q0 kek + kPd k 1 1X n
 2  4  q0 kek2 + kPk k2 + ~uT ~ui
+ S1 S2 + w2 + b1 + uT1 u1  Da yd + S1 e2 2 t i=1 i
  r h ~
  1
+ S1 eq1 + Dfq1 + dq1 + ~ uT1 u1 S1  Da u1  k1 S21 + S1 S2 + S1 w2 + S1 e2 + 1 u~T1 u1 + 1 d 1 d1
s1 s1 r1
1 1 Xn ð47Þ
 ~ d1 Da d1 + ~
uT ~ui ð40Þ
r1 t i=1 i According to (44) and (45), we have
Design the first intermediate control function b1 with 1 1X n

update laws D a V 1 4  q1 ke k2 + kPk k2 + ~uT ~ui

2 t i=1 i

b1 = k1 S1 + uT1 u1 ðx^1 Þ + signðS1 Þd1  Da yd r 1 ~T h 1

 k1 S21 + S1 S2 + u u1 + 1 ~d1 d1 + jS1 j2 + jw2 j2
ð41Þ s1 1 r1 2
a
D u1 = s1 u1 ðx^1 ÞS1  r1 u1 ð42Þ ð48Þ
Da d1 = r1 jS1 j  h1 d1 ð43Þ where q1 = q0  1=2.
By using DSC technique, we can obtain the state
According to Lemma 4, we have
variable v2 as
1 1
S1 e2 4 jS1 j2 + je2 j2 ð44Þ l2 Da v2 + v2 = b1 , v2 ð0Þ = b1 ð0Þ ð49Þ
2 2
1 1 according to (49), we have
S1 w2 4 jS1 j2 + jw2 j2 ð45Þ
2 2
Da w2 = Da v2  Da b1
Substituting(41), (42) and (43) into (40), we obtain
v 2  b1
=  D a b1 ð50Þ
1 1X n
l2
D a V 1 4  q0 k e k2 + kPk k2 + ~
uT ~
ui w2
2 t i=1 i = + B2
 l2
+ S1 S2  k1 S1  uT1 u1  signðS1 Þd1 + Da yd
   where B2 is a continuous function of variables
+ uT1 u1  Da yd + S1 w2 + S1 e2 + S1 eq1 + Dfq1 + dq1
  S1 , S2 , w2 , u1 , d1 , s3 , w3 , yd , Da yd , Da ðDa yd Þ.
~ T 1 Step 2. Define the error surface S3 and the output
+ u1 u1 S1  ðs1 u1 S1  r1 u1 Þ
s1 error of the fractional order filter
1~
 d 1 ðr1 jS1 j  h1 d1 Þ S3 = x^3  v3
r1 ð51Þ
w3 = v3  b2
1 1X n
4  q0 kek2 + kPk k2 + ~T u
u ~i
Further, we can obtain
2 t i=1 i
 k1 S21 + S1 S2  jS1 jd1 + S1 w2 + S1 e2 Da S2 = x^3 + uT2 u2 + ~uT2 u2 + k12 e1
  r T
+ S1 eq1 + Dfq1 + dq1 + 1 ~ u u1 + eq2 + Dfq2 + dq2  Da b1
s1 1 ð52Þ
h = S3 + w3 + b2 + uT2 u2 + ~uT2 u2 + k12 e1
~d1 jS1 j + 1 ~d1 d1 ð46Þ
r1 + eq2 + Dfq2 + dq2  Da b1
note that eq1 + Dfq1 + dq1 = D1 , we can obtain S1 D1 4 Construct the third Lyapunov function:
jS1 D1 j4jS1 jjD1 j4jS1 jd1 and jS1 jd1 = jS1 jð~d1 + d1 Þ.
and we have 1 2 1 ~T ~ 1 ~2 1
V2 = V1 + S2 + u2 u2 + d2 + w22 ð53Þ
2 2s2 2r2 2
Chen et al. 1251

we have Substituting equations (56)–(58) into (55), we obtain

1 ~T a ~ 1 Da V2
Da V2 4Da V1 + S2 Da S2 + u2 D u2 + ~d2 Da ~d2
s2 r2 1 1X n

+ w2 D w2 a 4  q2 ke k2 + kPk k2 + ~uT ~ui  k1 S2

1
2 t i=1 i
1
4  q1 kek2 + kPk k2  k1 S21 + S1 S2 + S2 ðS1 + S3  k2 S2  S1  uT2 u2  signðS2 Þd2
 2  
b1  v 2 b1  v 2  
+ S2 S3 + w3 + b2 + uT2 u2 + ~ uT2 u2  Da v2 + T ~ T
+u2 u2 + u2 u2  + jS2 j ~d2 + d2
  r T l2 l2
+ k12 S2 e1 + S2 eq2 + Dfq2 + dq2 + 1 ~ u u1 1 1
s1 1  ~uT2 ðs2 u2 S2 r2 u2 Þ  ~d2 ðr2 jS2 j  h2 d2 Þ
h 1 1X n s2 r2
+ 1~ d1 d1 + jS1 j2 + jw2 j2 + ~
uT ~
ui r h 1
r1 2 t i=1 i + 1 ~uT1 u1 + 1 ~d1 d1 + jS1 j2 + jS2 j2 + jw2 j2
s1 r1 2
1 ~T a ~ 1 w2 2
+ u2 D u2 + ~ d2 Da ~
d2  2 + B2 w2 1 w
s2 r2 l2 + jw3 j2  2 + B2 w2
2 l2
1 X
1 n ~T ~
4  q2 kek2 + kPk k2 + u ui  k1 S21 1 1X n
2 t i=1 i 4  q2 kek2 + kPk k2 + ~uT ~ui  k1 S2
1
 2 t i=1 i
+ S2 S1 + S3 + b2 + uT2 u2 + ~ uT2 u2 Da v2 Þ
 k2 S22 + S2 S3  jS2 jd2 + uT2 u2 S2
1 ~T a 1 w22   r
+ S2 D2  u D u2  ~ d2 Da d2  + jS2 j ~d2 + d2  ~uT2 u2 S2 + 2 ~uT2 u2  jS2 j~d2
s2 2 r2 l2 s2
r 1 ~T h1 ~ h r h
+ B2 w2 + u u1 + d1 d1 + jS1 j + jS2 j2
2
+ 2 ~d2 d2 + 1 ~uT1 u1 + 1 ~d1 d1 + jS1 j2 + jS2 j2
s1 1 r1 r2 s1 r1
1 1 1 1 w 2
+ jw2 j2 + jw3 j2 + jw2 j2 + jw3 j2  2 + B2 w2
2 2 2 2 l2
ð54Þ ð59Þ
2 2
where q2 = q1  k212 =2,
and k12 S2 e1 4jS2 j =2212 je1 j =2. By employing Young’s inequality, we have
From (49) we can obtain w2 B2
w2 B2 4 2 2 + 2m. Then, we have
1 1X n 2m
Da V2 4  q2 kek2 + kPk k2 + ~
uT ~ui  k1 S21
2 t i=1 i Da V2

b  v2 1 1X n
+ S2 uT2 u2 : 1
S1 + S3 + b2 + uT2 u2 + ~ 4  q2 ke k2 + kPk k2 + ~uT ~ui  k1 S2  k2 S2
1 2
l2 2 t i=1 i
  1 T a 1 r h r h
+ jS2 j ~d2 + d2  ~ u D u2  ~ d2 D a d2 + S2 S3 + 2 ~uT2 u2 + 2 ~d2 d2 + 1 ~uT1 u1 + 1 ~ d1 d1
s2 2 r2 s2 r2 s1 r1
w2 r ~T h ~ 1 1 w2 w2 B2
 2 + B2 w2 + 1 u u1 + 1 d 2 + jS1 j2 + jS2 j2 + jw2 j2 + jw3 j2  2 + 2 2 + 2m
1 d1 + jS1 j 2 2 l2 2m
l2 s1 1 r1
1 1 ð60Þ
+ jS2 j2 + jw2 j2 + jw3 j2
2 2 Step i. According to Theorem 1, design the error sur-
ð55Þ face Si + 1 and the output error of a fractional order
where eq2 + Dfq2 + dq2 = filter.
 D2 , S2 D2 4jS2 D2 j4jS2 jjD2 j, and
jS2 jjD2 j4jS2 jd2 4jS2 j ~d2 + d2 . Si + 1 = x^i + 1  vi + 1
Choose the first intermediate control function b2 ð61Þ
w i + 1 = vi + 1  b i
with update laws,
" # The intermediate control function bi and the update
k2 S2 + S1 + uT2 u2 + signðS2 Þd2 laws are designed as
b2 = ð56Þ
 b1lv 2
2 3
2
Si1 + ki Si + uTi ui + signðSi Þdi
Da u2 = s2 u2 ðx^1 , x^2 ÞS2  r2 u2 ð57Þ bi = 4 b  vi 5 ð62Þ
 i1
a
D d2 = r2 jS2 j  h2 d2 ð58Þ li
1252 Measurement and Control 54(7-8)

Da ui = si ui ðx^1 , x^1 , :::, x^i ÞSi  ri ui ð63Þ Da un = sn fn Sn  rn un ð73Þ

Da di = ri jSi j  hi di ð64Þ
Da dn = rn jSn j  hn dn ð74Þ
Employing DSC technique, vi can be obtain as
From (71)–(74), we obtain
li Da vi + vi = bi1 , vi ð0Þ = bi1 ð0Þ ð65Þ
X
n1
1X n
by (65), we have Da Vn 4  qn1 kek2  ki S2i + Sn1 Sn + ~
uT ~
ui
t i=1 i
wi i=1
Da wi = + Bi ð66Þ X
n1 X
n1 X
n1
li r h
+ wn Da wn + i ~
uTi ui + i ~
di di + jSi j2
a s
i=1 i
r
i=1 i
where Bi =  D ai1 . i=1
X
n1  
Consider Lyapunov function: 1 2
Bi 1 1 2 1
+ kPk k2 +  + wi + w2n
1 2 1 ~T ~ 1 ~2 1 2 i=2
2m li 2 2
Vi = Vi1 + S + u ui + d + w2i ð67Þ  T
2 i 2si i 2ri i 2 + 2mðn  2Þ + Sn u + wn + k1n e1 + un un
  
From (54) and (57), we obtain +~uT un  Da vn + Sn eq + Dfq + dq
n n n n
1 T 1
X  ~ un ðsn fn Sn  rn un Þ  ~ dn ðrn jSn j  hn dn Þ
i
1X n
sn rn
Da Vi 4  qi kek2  kk S2k + Si Si + 1 + ~
uT ~
ui
m=1
t i=1 i X
n1
1X n
4  qn1 kek2  ki S2i + Sn1 Sn + ~
uTi ~
ui
Xi
rk ~T Xi
hk ~ Xi
1 i=1
t i=1
+ uk uk + dk dk + jSk j2 + kPk k2
s
m=1 k m=1 k
r m=1
2 X
n1
r X
n1
h X
n1
  + wn Da wn + i ~
uTi ui + i ~jSi j2
di di +
Xi
B2k 1 1 2 1 i=1
s i i=1
r i i=1
+  + wk + w2i + 1 + 2mði  1Þ
2m lk 2 2 Xn1  2 
m=2 1 B 1 1 1
+ kPk k2 + i
 + w2 + w2n
ð68Þ 2 i=2
2m li 2 i 2

w2 B2 + 2mðn  2Þ + Sn Sn1  kn Sn  uTn un  signðSn Þdn
whereqi = qi1  k21i =2 and wi Bi 4 i i + 2m 
2m b  vn b  vn
+ n1 + uTn un  n1 +~ uTn un Sn
Step n. Define the error surface Sn and the output ln ln
error of a fractional order filter 1 ~T
+ Sn Dn + k1n Sn e1  u ðsn fn Sn  rn un Þ
sn n
Sn = x^n  vn 1~
ð69Þ  d n ðrn jSn j  hn dn Þ
wn = vn  bn1 rn
X
n1
1
The Lyapunov function is chosen as 4  qn kek2 + jSi j2 + kPk k2 + wn Da wn
i=1
2
1 2 1 ~T ~ 1 ~2 1 X
n1  
Vn = Vn1 + Sn + u un + d + wn 2 ð70Þ B2i 1 1 2 1
2 2sn n 2rn n 2 +  + wi + w2n + 2mðn  2Þ
i=2
2m l i 2 2
According to (68), we obtain X
n1
1X n
~
X
n1
ri ~T
1 ~T a ~  ki S2i + uTi ~
ui + ui ui
Da Vn 4Da Vn1 + Sn Da Sn + u D un i=1
t i=1 s
i=1 i
sn n
X
n1
h
1 + i~
di di  kn S2n  jSn jdn + ~uTn un Sn
+ ~ dn D a ~
dn + w n D a w n ri
rn i=1
 
Xn1
1X n + jSn j dn + ~ dn  ~ uTn un Sn
4  qn1 kek2  ki S2i + Sn1 Sn + ~
uTi ~
ui r ~T
t + nu un  jSn jd~n + hn d ~n dn
i=1 i=1
sn n rn
X
n1
r X
n1
h 1 X
n1
X n
1X n Xn
r i ~T
+ i ~
uTi ui + jSi j2 + kPk k2
i ~
di di + 4  qn kek2  ki S2i + ~
uTi ~
ui + u ui
s
i=1 i
r
i=1 i i=1
2
i=1
t i=1 s i
i=1 i
X
n1   X
n X
n
B2i 1 1 2 1 h 1
+  + wi + w2n + 2mðn  2Þ + i ~ jSi j2 + kPk k2 + wn Da wn
di di +
2m li 2 2 i=1
ri i=1
2
i=2
X
n1  
1 ~T a 1 B2i 1 1 2 1
+ Sn Da Sn  u D un  ~ dn Da dn + wn Da wn ð71Þ +  + wi + w2n + 2mðn  2Þ
sn n rn i=2
2m li 2 2
Design control law uðtÞ with update laws, ð75Þ
!
Sn1 + kn Sn + uTn un + signðSn Þdn where qn = qn1  k21n =2:
u = b  vn ð72Þ
 n1l
n
Chen et al. 1253

By using DSC technique, we can obtain Denote

n  
ln Da vn + vn = bn1 , vn ð0Þ = bn1 ð0Þ ð76Þ 1X ri T  h 2
j= ui ui + i d2 + 2 j Si j
2 i = 1 si ri i
and ð81Þ
1
Da wn = Da vn  Da bn1 + 2mðn  1Þ + kPk k2
2
vn  bn1 C = minf2qn1 =lmin ðPÞ, 2ki , hi ,
=  Da bn1 ð77Þ
ln    
2 1 B2i 1 ð82Þ
wn r i  si , 2  
= + Bn t li 2m 2
ln
where Bn =  Da an1 . 2 2 Then (80) becomes
w B
According to wn Bn 4 n n + 2m, we can have Da Vn 4  CVn + j ð83Þ
2m
X
n
1X
n X
n
r i ~T
Da Vn 4  qn kek2  ki S2i + ~uT ~ui + u ui According to (83), we can obtain
i=1
t i=1 i s i
i=1 i
X
n X
n Da Vn + MðtÞ = CVn + j ð84Þ
h 1
+ ~i
di di + jSi j + kPk k2
2

i=1
r i i=1
2 where MðtÞ50.
  X B2 1
n 1   According to Lemma 5, wo can obtain
wn i 1 2
+ wn  + Bn +  + w
ln 2m li 2 i jm
i=2
Vn 4Vð0ÞEa ðCta Þ + , ð85Þ
1 2 C
+ w + 2mðn  2Þ
2 n Then, we have
Xn
1X n Xn
r i ~T
4  qn kek2  ki S2i + ~uT ~ui +
i ui ui
t s jm
i=1 i=1 i=1 i limjVn ðtÞj4 ð86Þ
X
n
h X
n
1
t!‘ C
+ i~
di di + jSi j2 + kPk k2
r
i=1 i i=1
2 Since 12 jS1 ðtÞj2 4Vn ðtÞ, which yields that

X B2 1
n  rffiffiffiffiffiffiffiffi
i 1 2
+  + w + 2mðn  1Þ 2jm
2m li 2 i limjS1 ðtÞj4 ð87Þ
i=2 t!‘ C
ð78Þ
This means that if the parameters are suitable,
According to Lemma 4, we have kS1 ðtÞk can converge to an arbitrary small region in
t 2 ðt0 , ‘Þ. And it is easy to see that all signals will
1 T~ 1 remain bounded in a closed-loop system.
uTl ul 4  ~
~ u ul + uT u
2 l 2 l l ð79Þ
~ 1 2 1 2
dl dl 4  ~ dl + ~ dl Simulations
2 2
Substituting (79) into (78), we obtain In this section, we use the following two examples to
verify the validity of the proposed method. Figure 1
X
n
1X n
shows the block diagram of the designed control system.
Da Vn 4  qn1 kek2  ki z2i + ~
uT ~ui
i=1
t i=1 i
1X n
r i ~T ~ 1 X n
hi ~ 2 1X n
ri T 
Example 1
 ui ui  di + u u Consider a fractional order nonlinear system:40
2 i = 1 si 2 i = 1 ri 2 i = 1 si i i
 a
1X n
hi 2 X n
1 D x1 = x2 + fk1 ðXðtÞÞ + dk1 ðtÞ
+ di + jSi j2 + kPk k2 ð88Þ
2 i = 1 ri i=1
2 Da x2 = x1 + fk2 ðXðtÞÞ + dk2 ðtÞ
X n  
B2i 1 1 2 where a=0:98, fk1 = 0, f12 = x1 ðtÞ0:25x2 ðtÞ  x31 ðtÞ,
+  + w + 2mðn  1Þ
i=2
2m l i 2 i f22 = 2x1 ðtÞ  0:25x2 ðtÞ  0:5x21 ðtÞ, f32 = 4x1 ðtÞ  0:5x2
X n  
n
1X r i 2 ~T ~ ðtÞ  x31 ðtÞ, dk1 = 0, dk3 = 0:5 sinðtÞ and yd = sin t is
4  qn1 kek2  ki S2i   ui ui defined as the reference signal. The initial conditions
i=1
2 i = 1 si t
  are xð0Þ = ½0:1, 0:1T , x^ð0Þ = ½0:2, 0:2T , ui ð0Þ = ½0:01,
1 X hi ~ 2
n X B2 1
n
i 1 2
 di +  + w +j :::, 0:01T , di ð0Þ = 0:01 and choose design parameters
2 i = 1 ri 2m li 2 i
i=2 as k1 = 30, k2 = 20, si = 1, ri = 40, ri = 1, hi = 20,
ð80Þ select the gain matrix as K = ½100, 4000T .
1254 Measurement and Control 54(7-8)

Figure 1. The block diagram of the designed control system..

Figure 2. The trajectory of the fractional order Duffing- Figure 3. Signal tracking of x1 and tracking errors in example 1.
Holmes system.
of x1 and the comparison trajectories with traditional
Figures 2 to 7 show the simulation results of backstepping method. Figure 4 gives the trajectories of
Example 1. Figure 2 displays the trajectories of the estimation of x1 and its estimation error. Figure 5 gives
fractional order Duffing-Holmes system (88) with the trajectories of the x2 signal tracking and its estima-
a = 0:98. Figure 3 shows that the tracking trajectories tion. Figure 6 displays the trajectory of the control
Chen et al. 1255

Figure 4. The trajectories of x1 , ^x1 and estimation error in Figure 6. The control input in example 1.
example 1.

Figure 7. The trajectories of f ð xÞ and the switching signal sðtÞ

Figure 5. The trajectories of signal tracking of x2 and its
in example 1.
estimation in example 1.

where the system order is a = 0:97, the unknown func-

input. Figure 7 shows the trajectories of fðxÞ and the
tion is f1 = 5:5x1 ðtÞ  3:5x2 ðtÞ  0:8x3 ðtÞ  x31 ðtÞ,
switching signal sðtÞ. It can be seen from Figure 6 that
f = 2x1 ðtÞ  x2 ðtÞ  x3 ðtÞ  x31 ðtÞ, dk ðtÞ = 0:5 sinðtÞ is
2
fðxÞ changes randomly between fk ðxÞ. From the
external disturbance, yd = sin t is defined as the refer-
Example 1 simulation results, the proposed method can
ence signal.
obtain good control performance for fractional order
Choose the initial conditions of the system as
systems containing switched unknown functions and
xð0Þ = ½0:1, 0:1, 0:1T , the observer initial conditions
external disturbances.
are chosen as x^ð0Þ = ½0:2, 0:2, 0:2T and uð0Þ = ½0:01,
:::, 0:01T , dð0Þ = 0:1. Choose design parameters as
k1 = 20, k2 = 20, k3 = 40, s = 5, r = 40, r = 2,
Example 2
h = 40, select the gain matrix as K = ½15, 150, 1500T .
Consider the fractional order strict-feedback nonlinear Figures 8 to 14 show the simulation results of
system:39 Example 2. Figure 8 displays the trajectory of the
8 a fractional order Arneodo system (89) with a = 0:97.
>
> D x1 ðtÞ = x2 ðtÞ Figure 9 shows that the tracking trajectories of x1 and
< a
D x2 ðtÞ = x3 ðtÞ the comparison trajectories with traditional backstep-
ð89Þ
>
> Da x3 ðtÞ = u + fk ðXðtÞÞ + dk ðtÞ ping methods, it can be clearly seen from the figure that
:
y = x 1 ðt Þ the proposed method can converge faster and have a
1256 Measurement and Control 54(7-8)

Figure 8. The trajectory of the fractional-order Arneodo

system (89). Figure 10. The trajectories of x1 , ^x1 and estimation error in
example 2.

Figure 9. The trajectory of signal tracking of x1 and tracking

errors in example 2.

Figure 11. The estimation trajectories of x2 and x3 in example

smaller tracking error. It can be seen from Figure 10 2.
that the designed observer has good estimation perfor-
mance, and its estimation error can converge. Figure
11 gives the estimation trajectories of x2 and x3 , it can
obtain good estimation performance for x2 and x3 .
Figure 12 shows that the tracking trajectories of x2 and
x3 . Figure 13 displays the trajectory of the control
input. Figure 14 shows the trajectories of fðxÞ and the
switching signal sðtÞ of example 2.
From Example 1 and Example 2, we can conclude
that the proposed controller can stabilize the switched
fractional order systems of the equation (15), and the
control performance is satisfactory. The designed con-
trol law can ensure the convergence of the tracking
errors and the boundedness of all signals in the system.

Conclusions
This paper proposes an adaptive neural network back- Figure 12. The trajectories of signal tracking of x2 and x3 in
stepping sliding mode controller based on DSC to example 2.
Chen et al. 1257

Figure 13. The control input in example 2. Figure 14. The trajectories of f ð xÞ and the switching signal
sðtÞ in example 2.

stabilize the fractional order strict-feedback nonlinear

systems. The fractional order nonlinear system under
2. Gorenflo R and Mainardi F. Fractional calculus. Vienna:
consideration contains arbitrary switchings, unmea- Springer, 1997.
sured states, the unknown nonlinear functions and 3. Vinagre BM, Chen YQ and Petráš I. Two direct tustin
uncertain disturbances. The unknown nonlinear func- discretization methods for fractional-order differentia-
tions and uncertain disturbances are approximated by tor/integrator. J Franklin Inst 2003; 340(5): 349–362.
the radial basis function neural network and an obser- 4. Ionescu CM. The human respiratory system: an analysis
ver is designed for state estimation of the fractional of the interplay between anatomy, structure, breathing and
order systems. Fractional order DSC technology is used fractal dynamics. London: Springer Science & Business
to avoid ‘‘explosion of complexity’’ and obtain frac- Media, 2013.
tional derivatives for virtual control laws continuously. 5. Kilbas A, Srivastava HM and Trujillo JJ. Theory and
applications of fractional differential equations. North-Hol-
The stability of the closed-loop system is ensured by the
land mathematics studies, Vol. 204. Elsevier, 2006.
constructed Lyapunov functions. Examples and simula- Netherlands.
tion results show that the tracking error can quickly 6. Wang X. Mittag-leffler stabilization of fractional-order
converge to a small region of the origin by the proposed nonlinear systems with unknown control coefficients.
adaptive fractional order control method, and the semi- Adv Differ Equations 2018; 2018(1): 1–14.
global stability can be ensured in the closed-loop sys- 7. Ding D, Qi D, Peng J, et al. Asymptotic pseudo-state sta-
tem. An interesting future topic involves fractional bilization of commensurate fractional-order nonlinear
order systems Synchronization control, multi-agent sys- systems with additive disturbance. Nonlinear Dyn 2015;
tems control and fractional order model of DC model. 81(1): 667–677.
8. Li Y, Liu A and Yang C. Adaptive sliding mode control
Declaration of conflicting interests of a class of fractional-order nonlinear systems with
input uncertainties. IEEE Access 2019; 7: 74190–74197.
The author(s) declared no potential conflicts of interest with 9. Jain S, Hote YV, Dehuri P, et al. Experimental valida-
respect to the research, authorship, and/or publication of this tion of fractional order internal model controller design
article. on buck and boost converter. Meas Control. Epub ahead
of print 26 May 2020. DOI: 10.1177/0020294020922264.
Funding 10. Zhou J, Wen C, Wang W, et al. Adaptive backstepping
control of nonlinear uncertain systems with quantized
The author(s) received no financial support for the research,
states. IEEE Trans Autom Control 2019; 64(11): 4756–
authorship, and/or publication of this article.
4763.
11. Sun X, Yu H, Yu J, et al. Design and implementation of
ORCID iDs a novel adaptive backstepping control scheme for a
PMSM with unknown load torque. IET Electr Power
Tao Chen https://orcid.org/0000-0001-9596-8089
Appl 2019; 13(4): 445–455.
Jiaxin Yuan https://orcid.org/0000-0001-5394-7714
12. Cai J, Qian F, Yu R, et al. Adaptive backstepping control
for a class of non-triangular structure nonlinear systems.
References IEEE Access 2020; 8: 76093–76099.
1. Radwan AG, Ahmad TA, Sundarapandian V, et al. 13. Zhang J and Li J. Adaptive backstepping sliding mode
Fractional-order and memristive nonlinear systems: control for wheel slip tracking of vehicle with uncertainty
Advances and applications. Complexity 2017; 2017: 1–2. observer. Meas Control 2018; 51(9–10): 396–405.
1258 Measurement and Control 54(7-8)

14. Luo Y, Zhao S, Yang D, et al. A new robust adaptive systems. In: IEEE transactions on neural networks and
neural network backstepping control for single machine learning system, 22 October 2020, pp.1–10. New York:
infinite power system with tcsc. IEEE CAA J Autom Sin IEEE. America.
2019; 7(1): 48–56. 28. Lu S and Wang X. Observer-based command filtered
15. Wan M and Liu Q. Adaptive fuzzy backstepping control adaptive neural network tracking control for fractional-
for uncertain nonlinear systems with tracking error con- order chaotic PMSM. IEEE Access 2019; 7: 88777–88788.
straints. Adv Mech Eng 2019; 11(5): 1–11. 29. Tong S, Min X and Li Y. Observer-based adaptive fuzzy
16. Peng J and Dubay R. Adaptive fuzzy backstepping con- tracking control for strict-feedback nonlinear systems
trol for a class of uncertain nonlinear strict-feedback sys- with unknown control gain functions. IEEE Tran Cybern
tems based on dynamic surface control approach. Expert 2020; 50(9): 3903–3913.
Syst Appl 2019; 120: 239–252. 30. Tong S, Sui S and Li Y. Observed-based adaptive fuzzy
17. Yang W, Yu W and Zheng WX. Fault-tolerant adaptive tracking control for switched nonlinear systems with
fuzzy tracking control for nonaffine fractional-order full- dead-zone. IEEE Trans Cybern 2015; 45(12): 2816–
state-constrained miso systems with actuator failures. 2826.
IEEE Transactions on Cybernetic, 20 January 2021, pp.1– 31. Long L and Zhao J. Switched-observer-based adaptive
14. New York: IEEE. America. neural control of mimo switched nonlinear systems with
18. Fei J and Chen Y. Fuzzy double hidden layer recurrent unknown control gains. IEEE Trans Neural Networks
neural terminal sliding mode control of single-phase Learn Syst 2016; 28(7): 1696–1709.
active power filter. In: IEEE transactions on fuzzy sys- 32. Li Y and Tong S. Adaptive neural networks prescribed
tems, 29 July 2020, pp.1–1. New York: IEEE. America. performance control design for switched interconnected
19. Fei J, Chen Y, Liu L, et al. Fuzzy multiple hidden layer uncertain nonlinear systems. IEEE Trans Neural Net-
recurrent neural control of nonlinear system using terminal works Learn Syst 2017; 29(7): 3059–3068.
sliding-mode controller. In: IEEE transactions on cyber- 33. Li Y, Chen Y and Podlubny I. Mittag–leffler stability of
netics, 12 March 2021, pp.1–16. New York: IEEE. America. fractional order nonlinear dynamic systems. Automatica
20. Fei J, Wang Z, Liang X, et al. Fractional sliding mode 2009; 45(8): 1965–1969.
control for micro gyroscope based on multilayer recurrent 34. Podlubny I. Fractional differential equations. Elsevier,
fuzzy neural network. In: IEEE transactions on fuzzy sys- 1998. America.
tems, 9 March 2021, pp.1–1. New York: IEEE. America. 35. Duarte-Mermoud MA, Aguila-Camacho N, Gallegos
21. Liu H, Pan Y, Li S, et al. Adaptive fuzzy backstepping JA, et al. Using general quadratic Lyapunov functions to
control of fractional-order nonlinear systems. IEEE prove Lyapunov uniform stability for fractional order
Trans Syst Man Cyber Syst 2017; 47(8): 2209–2217. systems. Commun Nonlinear Sci Numer Simul 2015; 22(1–
22. Wang C and Liang M. Adaptive nn tracking control for 3): 650–659.
nonlinear fractional order systems with uncertainty and 36. Wang Y, Xie L and De Souza CE. Robust control of a
input saturation. IEEE Access 2018; 6: 70035–70044. class of uncertain nonlinear systems. Syst Control Lett
23. Ma Z and Ma H. Adaptive fuzzy backstepping dynamic 1992; 19(2): 139–149.
surface control of strict-feedback fractional-order uncer- 37. Li Y, Chen Y and Podlubny I. Stability of fractional-
tain nonlinear systems. IEEE Trans Fuzzy Syst 2019; order nonlinear dynamic systems: Lyapunov direct
28(1): 122–133. method and generalized mittag–leffler stability. Comput
24. Shahvali M, Naghibi-Sistani MB and Modares H. Dis- Math Appl 2010; 59(5): 1810–1821.
tributed consensus control for a network of incommensu- 38. Wang C, Liang M and Chai Y. An adaptive control of
rate fractional-order systems. IEEE Control Syst Lett fractional-order nonlinear uncertain systems with input
2019; 3(2): 481–486. saturation. Complexity 2019; 2019: 5643298.
25. Shahvali M, Azarbahram A, Naghibi-Sistani MB, et al. 39. Balootaki MA, Rahmani H, Moeinkhah H, et al. On the
Bipartite consensus control for fractional-order nonlinear synchronization and stabilization of fractional-order
multi-agent systems: an output constraint approach. Neu- chaotic systems: recent advances and future perspectives.
rocomputing 2020; 397: 212–223. Physica A 2020; 551: 124203.
26. Lu S and Wang X. Command filtering-based neural net- 40. Deepika D, Kaur S and Narayan S. Uncertainty and dis-
work control for fractional-order pmsm with input turbance estimator based robust synchronization for a
saturation. IEEE Access 2019; 7: 137811–137822. class of uncertain fractional chaotic system via fractional
27. Sui S, Chen CP and Tong S. Neural-network-based adap- order sliding mode control. Chaos Solitons Fractals 2018;
tive DSC design for switched fractional-order nonlinear 115: 196–203.