Abstract—The convergence analysis of two-dimensional
based integrated predictive iterative learning control (2D-
IPILC) is presented for batch process in presence of output
noises. In the 2D-IPILC method, iterative learning control (ILC)
in the batch domain is integrated reasonably with real-time
model predictive control (MPC) in the time domain. Based on
the 2D system theory, system response of output tracking error
is described explicitly by using the state-transition matrix of the
2D system for the first time. Then influence of output noise to
the tracking error can be revealed clearly. By using this
description of model response, convergence properties of the
2D-IPILC can be also analyzed theoretically. The sufficient
convergence conditions of output tracking error in the proposed
algorithm are derived for a class of linear systems. Simulation
results have demonstrated the effectiveness of the proposed
method.
I. INTRODUCTION
To deal with the tracking control problem in batch process,
iterative learning control (ILC) has become a popular control
framework [1,2]. ILC is applied widely to the batch processes
owing to its simple control scheme and intuitional
convergence condition [3]. Normally, zero tracking error in
ILC is very difficult to be achieved because the disturbances in
real industry process always exist. For example, the effect of
output noise on P-type ILC algorithm was analyzed by Owens
[4], and the conclusion has shown that the normal ILC
algorithms may converge to a bounded constant neighborhood.
Since the traditional P-type ILC is a kind of feed-forward
control framework, it is reasonable to introduce feedback
control into ILC for better performance while the output noise
exists. Recently, many researchers have focused on the
combination of ILC with model predictive control (MPC), and
it has been shown that such combination can achieve better
control performances [5-8].
However, it should be noticed that batch process can be
considered as a standard two-dimensional (2D) system [9], if
both batch domain and time domain are considered. More
research works have been presented since Kurek and Zaremba
[10,11] re-explained the traditional P-type ILC in the 2D
framework. Furthermore, a robust ILC scheme for batch
processes with uncertain perturbations was developed by
*This work was supported by the National Natural Science Foundation of
China, Grants 61473162, and EU FP7 Marie Curie International Research
Staff Exchange Scheme (Ref: PIRSES-GA-2013- 612230).
Z. H. Xiong, Y. D. Hong and C. Chen are with Department of Automation,
Tsinghua University, Beijing, China, 100084 (corresponding author Z. H.
Xiong: 86-10-62785845; fax: 86-10-62786911; e-mail: zhxiong@tsinghua.
edu.cn).
J. Zhang is with School of Chemical Engineering and Advanced Materials,
University of Newcastle, Newcastle upon Tyne, NE1 7RU, U.K (e-mail: jie.
zhang@ncl.ac.uk).
M. Y. Zhong is with College of Electrical Engineering and Automation,
Shandong University of Science and Technology, Qingdao, 266590, China.
using the linear matrix inequalities (LMI) technique [12].
Wang et al. [13] proposed an online iterative learning MPC
(ILMPC) for the multi-phase batch process control. Cueli and
Carlos [14] also proposed an iterative nonlinear MPC
(INMPC). To overcome the 2D interval repeatable and
unrepeatable uncertainties, the 2D dynamic matrix control
(2D-DMC) method was proposed and the properties of the
algorithm were also analyzed [15].
In spite of these works mentioned above, few works
involve the state-transition matrix and system responses in the
2D framework. In our previous work [16,17], the 2D based
integrated predictive ILC (2D-IPILC) has been proposed, and
it is illustrated on a typical batch reactor. However, the
convergence condition of 2D-IPILC was discussed intuitively
[17], and the robustness of the method should be analyzed
further in order to overcome the disturbances. In this paper,
the natural characteristics of 2D system, such as the state-
transition matrix and system responses, are adopted in the
algorithm, and the influence of all past excitation signals can
be interpreted clearly. As a result, the convergence condition
of output tracking error in the 2D-IPILC can be analyzed
theoretically. Therefore, sufficient convergence conditions of
output tracking error are derived for a general class of linear
systems without output noise. Even though zero tracking error
is very difficult to be reached in the presence of output noise, a
bounded value of 2D-IPILC can be reached.
The rest of the paper is organized as follows: the 2D-IPILC
algorithm is summarized in Section 2, and in Section 3, the
sufficient convergence conditions of output tracking error are
analyzed and proved theoretically. A numerical simulation is
presented in Section 4. Finally, conclusions are given in
Section 5.
II. SUMMARY OF 2D-IPILC
A. 2D System Model and ILC Law
A class of single-input single-output (SISO) linear time-
invariant (LTI) discrete-time system is considered here. It is
assumed that the system is consisted of N sampling steps in
each batch, and the system dynamics can be described by the
following discrete-time state-space model:
1, , ,
, , ,
x t k Ax t k Bu t k
y t k Cx t k d t k
where t and k represent the sampling time and the batch index,
xR
n
, uR, yR are state, input, and output variables,
respectively. A, B, and C are real matrices with appropriate
dimensions, and CB
0 is assumed here. d(t,k) denotes the
effect of disturbance, bias errors and measurement noise, and
d(t,k) is supposed here to be bounded by B
d
> 0, i.e. t,k,
Convergence Analysis of Integrated Predictive Iterative Learning Control
Based on Two-Dimensional Theory*
Zhihua Xiong, Yingdong Hong, Chen Chen, Jie Zhang, and Maiying Zhong
2016 American Control Conference (ACC)
Boston Marriott Copley Place
July 6-8, 2016. Boston, MA, USA
978-1-4673-8682-1/$31.00 ©2016 AACC 1259