Robust model predictive control of linear systems with constraints
Shuyou Yu, Yang Guo, Yu Zhou and Hong Chen
Abstract— In this paper, disturbance observer based model
predictive control of linear systems which satisfies a matching
condition is proposed, where the disturbance is bounded and
varying slowly. A conventional nominal model predictive control
problem with tightened constraints is solved online which
predicts the nominal trajectory. Two ancillary control laws
are determined off-line: one drives the trajectories of the real
system to the trajectories of the nominal system, the other
tries to cancel out the effect of the disturbance input. Both
recursive feasibility of the involved optimization problem and
robust stability of systems under control are guaranteed if
the optimization problem is feasible at the initial time instant.
The resultant online algorithm has similar complexity to that
required in conventional model predictive control.
I. INTRODUCTION
Model predictive control (MPC) is one of the most ef-
fective techniques available for the control of constrained
systems. At each time instant, an optimization problem is
solved with the measurement of the system state, and a
control sequence is obtained accordingly so as to predict
the systems dynamics in a time horizon. However, only is
the first segment of the sequence applied to the system. At
the next time instant, the whole procedure is repeated with
the updated measurement of the system state [1, 2].
While dealing with constrained systems with uncertainties,
an MPC algorithm is required to ensure the satisfaction
of all constraints at all times as well as to guarantee the
desired performance. It is natural to consider the worst-case
or min-max approach, in which a worst-case performance
is minimized online to obtain an optimal control action,
suppose that there is no other information of uncertainties
except for their bound [3–5]. In general, min-max MPC is
very conservative since the worst-case scenario has to be
considered at each time instant. On one hand, the worst-
case scenario may be not happen at all. On the other hand,
the computational burden is very heavy. A great deal of
effort is devoted to reduce the computational burden and the
conservativeness of the min-max MPC. For linear systems
with parameter perturbations or parameter uncertainties, min-
max cost function is replaced by its upper bound in [6–9],
where a cost function which is upper bound of the min-max
cost function is minimized, and a linear feedback control
law instead of a sequence of control action is adopted in
order to predict the dynamics of the systems. The involved
optimization problem solved online can be reduced to a semi-
definite programming problem. The idea is easily extended
Shuyou Yu, Yang Guo, Yu Zhou and Hong Chen are with State Key
Laboratory of Automotive Simulation and Control, and with Department
of Control Science and Engineering, Jilin University, Changchun 130025,
P. R. China (e-mail: {shuyou,chenh}@jlu.edu.cn)
to linear parameter varying (LPV) systems [10, 11]. For
systems with exogenous disturbances, tube MPC or MPC
with tightened constraints are introduced in [12–15], where
the control action consists of a nominal control action and a
control law. The control action which is obtained by solving
online a nominal optimization problem drives the systems
dynamics of the nominal systems to the equilibrium. The
control law which is obtained offline drives the actual system
dynamics to the nominal system dynamics. Tube MPC has
almost the same computational burden compared with the
nominal MPC, but it only fits for restricted class systems
such as linear systems [12, 13] or Lipschitz nonlinear systems
[14, 15]. Furthermore, it cannot achieve satisfying effects in
controlling systems in the presence of strong disturbances.
Disturbance observer based control can handle the distur-
bances directly in the process of controller design, rather
than asymptotically suppress disturbances through feedback
regulation [16–18]. Compound control schemes combining
a feedforward compensation part based on disturbance ob-
server and a feedback regulation part based on MPC are
addressed to improve disturbance rejection performance of
the systems under control [19, 20], where control of both ball
mill grinding circuits and dead-time processes is considered,
respectively. Robust MPC with a disturbance observer for
three-phase voltage source PWM rectifier is presented in
[21]. The proposed method has an inherent rapid dynamic
response as a result of the conventional MPC. Offset-free
MPC is proposed in [22] where the objective of offset-
free is achieved by synthesizing an observer for the nom-
inal systems. Robust MPC for multivariate ill-conditioned
systems is addressed in [23], which shows that the optimal
disturbance model is close to the input disturbance model.
An explicit nonlinear MPC and disturbance observer based
control for trajectory tracking of autonomous helicopters
is introduced in [24], which provides an effective way of
integrating disturbance information.
In this paper, robust MPC of linear systems satisfying a
matching condition is considered, where the disturbances are
varying slowly and the bound of disturbances could be large.
A linear disturbance observer is designed offline to eliminate
or cancel off the influence of disturbances. The same as
tube MPC, only is a nominal optimization problem solved
online. Thus, the proposed scheme has a mild computational
burden in general. Both the recursive feasibility of the
optimization problem and the ultimate bound of the systems
under control are guaranteed suppose that the optimization
problem is feasible at the initial time instant. While either the
disturbances or derivative of the disturbances are decaying,
the system dynamics will approach to the equilibrium.
2017 11th Asian Control Conference (ASCC)
Gold Coast Convention Centre, Australia
December 17-20, 2017
978-1-5090-1572-6/17/$31.00 ©2017 IEEE 928