
www.ietdl.org
Published in IET Control Theory and Applications
Received on 28th February 2014
Revised on 13th May 2014
Accepted on 6th June 2014
doi: 10.1049/iet-cta.2014.0205
ISSN 1751-8644
Sampled-data MPC for LPV systems with input
saturation
Ting Shi, Hongye Su
State Key Laboratory of Industrial ControlTechnology, Institute of Cyber-Systems and Control, Zhejiang University,
Yuquan Campus, Hangzhou 310027, People’s Republic of China
E-mail: tingshi@hdu.edu.cn
Abstract: In this work, a sampled-data model predictive control (MPC) design method is proposed for continuous-time linear
parameter varying (LPV) systems. The input saturation and parameter uncertainties are both considered. Using a method
to deal with actuator saturation, the MPC controller is permitted to saturate. Using the measurable parameter vector, a
scheduled state-feedback MPC controller is computed at each time instant which fully exploits the real-time information on
the variations of the plant characteristics. By modelling the closed-loop systems of the continuous-time LPV systems with a
piecewise constant sampled-data control input as linear impulsive systems, the stability properties of the proposed MPC are
studied. The sampling interval in this work is not required to be periodic. The proposed MPC design method is expected to
further reduce the conservativeness. The improvements of the proposed sampled-data MPC method w.r.t. other existing MPC
techniques are demonstrated by an example.
1 Introduction
Over the past years, the wide popularity of model predictive
control (MPC) in both academia and industrial applications
has motivated the development of robust MPC techniques,
capable of dealing with control constraints and model uncer-
tainties; see, for example, [1–3] and references therein. Input
constraints are common encountered in practical control sys-
tems that arise from physical and technological constraints.
One approach of MPC to deal with input constraints is called
the low-gain control approach. See, e.g. [1–3]. Using this
approach, the control limits are never reached, that is, the
designed controller is not allowed to saturate. This type of
controller is in essence a linear one. It is well known that
low-gain controllers that avoid control limits often result in
low levels of performance [4]. Another method dealing with
input constraints is the use of saturated control. See, e.g.
[4–6]. Using this method, the proposed MPC design algo-
rithms permit the controller to saturate. Hence, it can lead
to the reduction of the conservativeness. In this work, a new
method proposed in [6] is used to deal with input constraints.
Compared with the method used in [4, 5], this one is much
simpler and can also obtain the saturated controller.
In recent years, the linear parameter varying (LPV) sys-
tem is the most popular model for MPC control (see, e.g.,
[7–11]). LPV system can be usually described by a polytopic
family of the linear system, parameterised by a time-varying
parameter vector, which is restricted to lie into a unit sim-
plex. The time-varying parameter is available online and
thus provides real-time information on the variations of the
plant characteristics. Hence, it is desirable to design MPC
controllers that are scheduled based on this information.
Early attempts of exploiting MPC for LPV systems use
the quadratic stabilisation paradigm, e.g., [1]. In this case,
the LPV systems are considered as uncertain polytopic sys-
tems. Then a single linear controller is designed for the
whole parameter uncertainties. This approach can lead to
conservativeness results. To reduce the conservativeness,
the class of linear parameter-dependent Lyapunov functions
has been introduced (see [9, 10]). Then, a linear scheduled
controller is computed at every time instant. The control
performance improvements are remarkable at the price of an
increased number of LMI conditions. More recently, in [11],
the special class of non-linear parameter-dependent Lya-
punov functions is used and a non-linear scheduled control
law is obtained which leads to further improve the control
performance.
However, most of the robust MPC algorithms available
nowadays are based on a discrete-time formulation of the
plant under control, while in the industrial process it is
often preferred to resort to a continuous-time description
of the system [12]. This is because that the plant model
is usually derived by resorting to first principles equations.
Then a continuous-time representation is much more natu-
ral. Some MPC algorithms for continuous-time systems have
been proposed in [12–16]. However, in these works, the
sampling intervals are assumed to be constant. It is hard
to perform constant sampling in some modern applications.
For instance, resources for measurement and control input
are restricted in networked systems. In this case, communi-
cation delays, package losses and heavy temporary load of
computation in a processor can lead to significant variations
IET Control Theory Appl., 2014, Vol. 8, Iss. 17, pp. 1781–1788 1781
doi: 10.1049/iet-cta.2014.0205 © The Institution of Engineering and Technology 2014