T. SHI, Z.-G. WU AND H. SU
parameter uncertainties and disturbance. In [21], a new type of dynamic output feedback RMPC is
presented for linear systems. With some controller parameters given in advance, more degrees of
freedom are provided. This leads to the improvement of the designs.
The RMPC designs usually use the measured states as initial condition to predict the future sys-
tem behavior. The main difficulty in the output feedback RMPC design is that the true states are not
available. In this case, a set should be used to obtain an estimation. In [17, 18], the estimated set is
fixed at every time instant. This is equivalent to imposing some additional constraints to the opti-
mization problem, which can lead to conservative results. In [19–21], the estimated set is refreshed
at every time instant. With the new estimated set, however, the RMPC optimization problem is
changed, and thus, the recursive feasibility can not be guaranteed directly. Then, some additional
conditions are used. In this work, with a modified RMPC design requirement and an additional con-
dition, the recursive feasibility can also be guaranteed. In comparison with [19–21], less conditions
are used. This can lead to the reduced computation burden.
Moreover, input constraints are commonly encountered in practical control systems that arise
from physical and technological constraints. Roughly speaking, there are two common methods to
deal with input constraints. One is the low-gain feedback approach, that is, the designed feedback
gains are sufficiently low such that the control limits are never reached (e.g., [17–20]). Using this
approach, the control inputs are not permitted to saturate. Thus, they are subject to more stringent
constraints than they should be. Another way is the use of saturated control input, for example, [7, 8]
for state feedback and [21] for output feedback RMPC. This approach permits the control inputs to
saturate and thus can fully utilize the capability of actuators. Compared with the low-gain feedback
approach, the second way enables us to fully utilize the capability of actuators and then to improve
the control performance. In this work, the second way is used to deal with input constraints.
In this work, a new algorithm of output feedback RMPC for linear uncertain systems with input
constraints is proposed. Compared with the most existing algorithms, our contribution is most sim-
ilar to [21]. As in [21], a saturated control signal is applied to the system, which is permitted to
saturate and thus can fully utilize the capability of the actuators. An ellipsoidal set is used to bound
the estimation error, and it is updated at every time instant. Then, some additional conditions are
used to guarantee the recursive feasibility. The main differences to [21] are as follows. First, some
controller parameters in [21] are given in advance and fixed, while in our work all the controller
parameters are online optimized. Thus, our algorithm can simplify the design procedures. Second,
with a modified RMPC design requirement, only one additional condition is used to guarantee the
recursive feasibility and asymptotic stability. In comparison with [21], less conditions are used, and
the computation time can be reduced. These are the main contributions of this work.
This paper is organized as follows. Section 2 gives the problem formulation and some prelimi-
naries. The design method of this new dynamic output feedback RMPC is given in Section 3. The
effectiveness of the proposed approach is illustrated in Section 4, and a conclusion in Section 5 ends
this paper.
Notation The notation is standard. I is the identity matrix with appropriate dimensions. The ele-
ments of a matrix A 2 R
nm
are denoted by A
.i;j /
;iD 1;:::;n; j D 1;:::;m and A
.i/
denotes the ith row of matrix A. For vector x, kxk
2
W
;W > 0 denotes its weighted vector 2-
norm, that is, kxk
2
W
D x
T
Wx. x.kjk/ or x.k/ denotes the state measured at real time k. Denote
x.k C ijk/ and u.k C ijk/ as the system state and control input of time k C i, predicted at time
k. For a positive-definite matrix P 2 R
nn
, denoted as P>0, a matrix H 2 R
mn
and a scalar
>0denote the ellipsoid
E .P; / , ¹´ 2 R
n
W ´
T
P´ 6 º and define the polyhedral set
L.H /
4
D
®
´ 2 R
n
WjH
./
´j 6 1; D 1;:::;m
¯
. E .P; 1/ is also denoted by E .P / for simplicity.
For a symmetric matrix, denotes the entries implied by matrix symmetry.
2. PROBLEM FORMULATION AND PRELIMINARIES
Let us consider the following uncertain system
x.k C 1/ D A.˛.k//x.k/ C B.˛.k//u.k/;
y.k/ D C.˛.k//x.k/;
(1)
Copyright © 2015 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control (2015)
DOI: 10.1002/rnc