Data-Driven based Iterative Learning Control for A Class of
Discrete-Time Descriptor Systems
Daqing Zhang
1
, Jie Yu
1
Baoyan Zhu
2
,
1. School of Science, University of Science and Technology Liaoning, Anshan, Liaoning Province, 114051, China
E-mail: d.q.zhang@ustl.edu.cn,
E-mail: yujie
ustl@sina.com
2. School of Science, Shenyang Jianzhu University, Shenyang, Liaoning Province, 110168, China
E-mail: zby1109@163.com
Abstract: Data-driven iterative learning control (ILC) for discrete-time descriptor systems is concerned. The input-output
property of single-input and single-output (SISO) discrete-time descriptor system is analyzed firstly. Then the relative degree of
the underlying descriptor is investigated. Under the assumption that the system is causal, an ILC algorithm is presented based
on the lifted form of descriptor system with zero relative degree. The ILC algorithm update the inputs serial by using the system
tracking error, and does not need to know the inner structure of the target descriptor system in advance. Simulation results shows
that, the presented ILC algorithm can make the target descriptor system’s output track the desired output well.
Key Words: Descriptor systems, Iteration Learning Control, Data-driven, Tikhonov regularization, Norm-optimal
1 Introduction
Descriptor systems, which are also named as differen-
tial algebra systems, singular systems, or semi-state sys-
tems, are found in the area of aeronautics and astronautics,
robots, power systems, electrical networks, chemical indus-
try, bioengineer- ring and economics [1, 2]. After more than
three decades researches, many excellent results on the de-
scriptor system have been obtained. Up to now, research on
this type system is still a hot branch in the control theory and
control engineering [3–8].
Iterative learning control (ILC) is based on the notion that
the performance of a system that executes the same task
multiple times can be improved by learning from previous
executions (trials, iterations, passes) [9]. Data-driven ILC
algorithms [10] estimate a system representation using in-
put/output data obtained during the ILC procedure and hence
avoid time-consuming identification experiments prior to the
ILC procedure.
As to the studies on ILC for descriptor systems, there are
some results can be accessed[11–20]. However, the existing
results of ILC for descriptor systems are all based on the
analysis of the inner structure of the descriptor systems.
In this paper, the data-driven ILC problem for descriptor
system is concerned. That is the inner structure of the target
descriptor system is supposed unknown, and only the input
and output data can be recorded.
The remainder of the paper is organized as follows. In
section 2, the relative degree of SISO discrete-time descrip-
tor system is investigated. The estimation of the convolution
matrix of the system is discussed in section 3. In section
4, the norm-optimal ILC algorithm is presented for SISO
discrete-time descriptor systems. Simulation is carried out
in section 5, and the paper is summarized in section 6.
This work is supported by National Natural Science Foundation (NNS-
F) of China under Grant 61273011,61273003.
2 Discrete-time descriptor systems’ relative de-
gree
Consider a SISO discrete-time linear descriptor system
Ex(k +1)=Ax(k)+Bu(k) (1a)
y(k)=Cx(k) (1b)
where, x
k
∈ R
n
is the state. E, A, B, C are system matrices
with appropriate dimensions. The matrix E is singular, e.g.
Rank(E) <nin general. The descriptor system (1) is said
to be regular if there is some λ in complex plane such that
det(λE − A) =0, causal if deg det(λE − A)=Rank(E),
and stable if the finite general eigenvalues of (E, A) are all
lie in the unit circle of the complex plane [1, 2]. For a given
descriptor system (1a) and (1b), there always exist a non-
singular matrices H and Q such that the system is restricted
system equivalent to
I 0
0 N
x(k +1)=
A
1
0
0 I
x(k)+
B
1
B
2
u(k) (2a)
y(k)=
C
1
C
2
x(k) (2b)
where, x(k)=H
x
1
(k) x
2
(k)
T
, x
1
(k) ∈ R
r
,x
2
(k) ∈
R
n−r
. N is a nilpotent matrix with N
h−1
=0and N
h
=
0. Here, the index h of the nilpotent matrix N is also the
index of the descriptor system (1a) and (1b). When h ≤ 1
the system is causal, otherwise, the system is not a causal
system.
In (2a), the system is divided into two subsystems, name-
ly, the forward recurrent subsystem
x
1
(k +1)=A
1
x
1
(k)+B
1
u(k) (3)
and backward recurrence
Nx
2
(k +1)=x
2
(k)+B
2
u(k) (4)
For k =0, 1, ··· ,L, the solution of the forward subsys-
tem is [1]
x
1
(k)=A
k
1
x
1
(0) +
k−1
i=0
A
k−i−1
1
B
1
u(i) (5)
Proceedings of the 35th Chinese Control Conference
Jul
27-29, 2016, Chen
du, China
3178