A Novel Data-driven Predictive Control for Networked Control Systems
with Random Packet Dropouts
Shuo Zhen, Zhongsheng Hou*, Chenkun Yin
Advanced Control Systems Laboratory, School of Electronics and Information Engineering, Beijing Jiaotong University, Beijing 100044
E-mail: 14120220@bjtu.edu.cn, zhshhou@bjtu.edu.cn, chkyin@bjtu.edu.cn
Abstract: A novel data driven predictive control method for networked control systems (NCSs) is presented, where the network
links from sensors to controllers and from controllers to actuators are subject to random packet dropouts. In order to make full use
of the I/O data of the controlled system to improve system performance, an improved model-free adaptive predictive control
(iMFAPC) method is proposed by modifying the criterion function. Then, a networked control scheme is developed based on
iMFAPC, whose basic principle is to compensate the lost packets by the corresponding predictive values. Finally, the
effectiveness of the proposed control scheme for the packet dropout problem in the networked control system is validated
through both simulations and experiments.
Key Words: Data-driven Control; Networked Control System; Random Packet Dropouts; Model-free Adaptive Predictive
Control
1 Introduction
With the development of control science, networked
control systems (NCSs) have been applied in different kinds
of fields. NCSs are distributed real-time feedback control
systems in which the control loop are closed via
communication networks [1]. Communication networks
make it possible to transfer data among different network
nodes, reduce the complexity in wired connections and
maintain control systems easily [2, 3]. However, the
insertion of communication networks causes many new
problems inevitably, such as packet dropouts, measurement
quantization, network-induced time delay etc.
Packet dropouts and network-induced time delay are two
primary factors which can deteriorate system performance
or even destabilize the control system. Therefore, both
packet dropouts and network-induced time delay are
research focuses. In practice, most of the industrial
processes dynamics are slow enough to ignore the
network-induced time delay. In this situation, the study on
packet dropouts is more significant than network-induced
time delay [4].
In general, the mainstream approaches dealing with
packet
dropout problem can be divided into two major
categories. The first category is to build a model for the
generalized controlled system including the packet dropout
process. In [5], a class of rate-constrained asynchronous
dynamic systems is derived. The quadratic stability is
analyzed by linear matrix inequality (LMI) method. In [6],
the arbitrary packet dropout process and the Markov packet
dropout process are analyzed. The corresponding controller
design techniques are developed based on Lyapunov
stability. In [7], the NCS with both packet dropouts
network-induced time delay is described as a class of novel
linear switched system. And the quantitative relationship
between the stability and the packet-dropout rate can be
established through the exponential stability condition for
*
This work is supported by National Natural Science Foundation
(NNSF) of China under Grants 61433002 and 61403025
*Corresponding author
the closed-loop system. In [8], the NCS with random packet
dropouts is modelled as a class of discrete-time linear
systems with Markovian jumping parameters. The
necessary and sufficient conditions for the robust H
∞
controller is given. In [9], LQG optimal control problem for
discrete-time NCS is analyzed. The critical value of the
packet-dropout rate is obtained, under which the NCS can
be stabilized. In [10], the process of packet dropout is
described as a Bernoulli process. The finite and infinite time
horizon optimal control problems of the linear
time-invariant system with TCP and UDP protocol are
studied respectively.
The other category dealing with packet dropouts is to
compensate the lost packets by predictive values. In [4], an
linear interpolation algorithm is used to predictive the
control output in the future time instants. In [11], a robust
model predictive control (MPC) method is proposed for the
linear time-invariant NCS with data quantization and packet
dropouts problem. The exponential stability of the
closed-loop system was derived by Lyapunov approach. In
[12], a novel networked predictive control (NPC) scheme is
proposed to deal with network-induced time delay and
packet dropouts. The stochastic stability criterion of
closed-loop system is presented.
The aforementioned methods are mostly designed for
linear systems, and require accurate mathematical models.
However, most industrial processes are too complex to
establish their accurate mathematical models [13].
Moreover, it becomes more difficult to build accurate
system models after inserting the communication network in
NCS. More complex factors should be considered. The
emerging data-driven control methods provide another way
for the study of NCS. Data-driven control methods depend
only on the I/O data of the controlled system [14]. In [15], a
data-driven control scheme based on subspace projection
method is used to produce the predictive outputs to
compensate for the effect of packet dropouts and
network-induced delay. In [16], a new data-driven control
method called data-based networked predictive control
(DBNPC) is developed for the NCS with packet dropouts.
2017 IEEE 6th Data Driven Control and Learning Systems Conference
Ma
26-27, 2017, Chon
qin
, China
978-1-5090-5461-9/17/$31.00 ©2017 IEEE
DDCLS'17
335