CNN-Based Adaptive Tracking Control for High
Speed Train with Asymmetric Input Saturation
Xianggui Guo, Junjie Zhao and Wendong Xiao
School of Automation and Electrical Engineering, University of Science and Technology Beijing
Beijing Engineering Research Center of Industrial Spectrum Imaging
Key Laboratory of Knowledge Automation for Industrial Processes, Ministry of Education
Beijing 100083, China
Email: guoxianggui@ustb.edu.cn
Abstract—This paper studies the adaptive position and velocity
tracking control for high speed train with asymmetric nonlinear
input saturation. Adaptive tracking control for high speed train
encounters great challenges when the train dynamics contain-
ing nonlinearities, asymmetric input saturation and unknown
external disturbances. An adaptive tracking control method is
developed by means of sliding mode technique combining with
adaptive Chebyshev neural network (CNN) technique. Through
rigorous analysis, it is concluded that the design control method
can ensure uniformly ultimate bounded convergence of train
trajectory to the reference profile. In particular, the assumption
with zero initialization condition for train velocity and reference
velocity is removed. Numerical simulation results further verify
the effectiveness of the proposed method.
Keywords —Adaptive tracking control; High speed train;
Chebyshev neural network (CNN); Sliding mode control; Asym-
metric input saturation
I. INTRODUCTION
High speed train with more efficient, rapidity and economy
has recently attracted a considerable research and attention
[1]–[5], especially China. As a practical application system,
the railway system is inevitably nonlinear subject to the
aerodynamic problems and the physical limitation of actua-
tors [2], [3]. This makes excellent tracking performance be
difficult to achieve. Due to some physical limitations of the
ctuators/servovalve, sensors and electromagnetic damping, the
actuator saturation problem is unavoidable during the train
motion control [3], which often severely degrades the system
performance, give rise to undesirable inaccuracy and even
instability [6], [7]. Although the high speed railway systems
with actuator saturations have recently been investigated ex-
tensively [2]–[4], asymmetric nonlinear actuator saturation is
not considered. Since the saturation in the high speed rail-
way actuators is very complicated with multivalues (traction
or braking operations) and non-smooth features, asymmetric
nonlinear actuator saturation, which first is introduced in [6],
can better character the actual physical constraints than those
considered in [2]–[4]. However, the nonlinear functions of the
asymmetric actuator saturation in [6] are assumed to be strictly
monotonous. This obviously limits its application in the real
systems. Therefore, how to remove this assumption motivates
the work of this paper.
Nonlinear dynamics often exist in high speed railway sys-
tems due to air-drag force nonlinearity, driving force con-
straint, braking force limitation and environment disturbances
such as wind gusts [5], [8]. It is well known that adaptive
sliding-mode control has the advantage of combining the
robustness of the sliding mode control with the tracking
abilities of the adaptive control [9]–[12]. Therefore, it is an
effective robust control approach for the nonlinear systems,
such as the high speed railway systems. On the other hand,
Chebyshev neural network (CNN) with superior approximated
ability is extensively applied to estimate the unknown lumped
uncertainty online to arbitrary accuracy [13], [14]. Therefore,
we are interested in investigating the position and velocity
tracking control problem for high speed train by combining
adaptive sliding mode control and CNN approach to improve
the control performance of high speed railway systems from
different aspects.
Motivated by the above discussion, the problem of adaptive
position and velocity tracking control for high speed railway
systems is investigated. The train motion dynamics containing
nonlinearities, asymmetric nonlinear actuator saturation and
unknown external disturbances are formulated as a nonlin-
early parameterized system. Adaptive sliding mode control
combining with CNN technique is proposed to guarantee
the uniform ultimate bounded convergence of train speed to
the reference profile. The assumptions of zero initialization
condition for train velocity and reference velocity in [3]
and strictly monotonous of the nonlinear functions in the
asymmetric actuator saturation as in [6] are removed.
Throughout this paper, the following notations are used: 1)
Let ℛ denote real numbers, ℛ
𝑛
denote the real 𝑛 vector, and
ℛ
𝑛×𝑚
denote the real 𝑛 ×𝑚 matrices; 2) ∥⋅∥stands for the
Euclidean norm of a vector; 3) The symbol of ∣⋅∣represents
the absolute value of real numbers; 4) 𝑡𝑟{𝑋} denotes the trace
of the matrix 𝑋.
II. H
IGH SPEED TRAIN MODEL AND PRELIMINARIES
In this section, combining tractive efforts, resistances and
train masses, the dynamic model of the high speed train is
presented and some preliminaries are also presented.
Eighth International Conference on Intelligent Control and Information Processing
November 3-5, 2017; Hangzhou, China
978-1-5386-1168-5/17/$31.00 ©2017 IEEE