J. Shanghai Jiaotong Univ. (Sci.), 2014, 19(6): 1-8
DOI: 10.1007/s12204-009-0501-3
Distributed Model Predictive Control with One-Step Delay
Communication for Large-Scale Systems and a Case Study
ZHANG Yan
∗
(张 艳), XU Cheng (徐 成)
(Logistics Engineering College, Shanghai Maritime University, Shanghai 201306, China)
© Shanghai Jiaotong University and Springer-Verlag Berlin Heidelberg 2014
Abstract: A distributed model predictive control (MPC) scheme with one-step delay communication is proposed
for on-line optimization and control of large-scale systems in this paper. Co operation between subsystems is
achieved by exchanging information with neighbor-to-neighbor communication and by optimizing the local prob-
lem with the improved performance index in the neighborhood. A distributed MPC algorithm with one-step delay
communication is developed for the situation that there is a one-step delay in the information available from its
neighbors when a subsystem solves the local optimization problem. The nominal stability is employed for the
whole system under the distributed MPC algorithm without the inequality constraints. Finally, the case study of
the reactor-storage-separator (RSS) system is illustrated to test the practicality of the presented control algorithm.
Key words: large-scale systems, model predictive control (MPC), exchange information, one-step delay commu-
nication, reactor-storage-separator (RSS)
CLC number: TP 273 Document code: A
0 Introduction
A large-scale system is often considered as a set of in-
terconnected subsystems, such as power systems, com-
puter and telecommunications networks, economic sys-
tems and multi-agent systems. The technical target is
to achieve some global performance of entire system.
The classical centralized control solution is often im-
practical to apply to large-scale systems owing to the
complexity of control synthesis and the physical restric-
tions on information exchange among subsystems
[1]
. It
is often required to adopt distributed or decentralized
framework where each subsystem is controlled by an
independent controller depending only on local mea-
surements, which has the advantage of being flexible
to system structure, error-tolerance and less computa-
tional efforts. With the development of distributed con-
trol system (DCS), field-bus and communication net-
work technologies in process industries, the distributed
control framework is usually adopted in large-scale sys-
tems.
Model predictive control (MPC), also called reced-
ing horizon control (RHC) or moving horizon control
Received date: 2013-10-30
Foundation item: the National Natural Science Founda-
tion of China (No. 61203110), the Shanghai Natural Sci-
ence Foundation (No. 13ZR1418900), and the Innova-
tion Programs of Shanghai Municipal Education Com-
mission (Nos. 12ZZ155 and 14YZ107)
∗E-mail: zhangyan@shmtu.edu.cn
(MHC), is widely recognized as a high practical con-
trol technology with high performance employed to-
day in the process industries, especially in the chem-
ical industry
[2-4]
. The ability to incorporate complex
objectives as well as constraints in a unified framework
makes it an extremely attractive and handy tool. Com-
pletely centralized MPC framework of large-scale sys-
tems is impractical. Completely decentralized control
of such systems, on the other hand, frequently results
in unacceptable control performance due to the inter-
connections between the subsystems ignored. Nowa-
days, distributed MPC is also gradually developing for
the control of large-scale systems. Some researches on
distributed MPC are available in the literatures
[5-10]
.
Among them, the distributed method described in
Ref. [5] is proposed for a set of decoupled subsystems,
which could handle systems with weakly interacting
subsystem dynamics. The iterative, cooperative dis-
tributed MPC algorithms with neighborhood optimiza-
tion were presented in Refs. [6] and [7] for large-scale
processes in which subsystems are interconnected by
inputs and by both inputs and states, respectively. Ref-
erence [8] prop osed an iterative, cooperative distributed
linear MPC strategy in which the subsystem controllers
optimize the same objective function in parallel without
the use of a coordinator, which guarantees performance
improvement and the appropriate communication bur-
den among subsystems. However, it is possible for each
subsystem to exchange the information several times