xvi Preface
system is not necessarily stable, and if it is too large, the predictive control
system will encounter a numerical stability problem. To overcome this prob-
lem, in Chapter 4, we propose the use of an exponentially weighted moving
horizon window in model predictive control design. Within this framework,
the numerical ill-conditioning problem is resolved by a simple modification
of the design model. Equally important are an asymptotic stability result
for predictive control designed using infinite horizon with exponential data
weighting and a modification of the weight matrices, as well as a result that
establishes the predictive control system with a prescribed degree of stability.
Analytical and numerical results are used in this chapter to show the equiv-
alence between the new class of discrete-time predictive control systems and
the classical discrete-time linear quadratic regulators (DLQR) without con-
straints. When constraints are present, the optimal solutions are obtained by
minimizing the exponentially weighted cost subject to transformed inequality
constraints.
Chapters 6 to 8 will introduce the continuous-time predictive control re-
sults. To prepare the background, in Chapter 5, we will discuss continuous-
time orthonormal basis functions and their applications in dynamic system
modelling. Laguerre functions and Kautz functions are special classes of or-
thonormal basis functions. Both sets of functions possess simple Laplace trans-
forms, and can be compactly represented by state-space models. The key prop-
erty is that when using the orthonormal functions, modelling of the impulse
response of a stable system, which has a bounded integral squared value,
will have a guaranteed convergence as the number of terms used increases.
This forms the fundamental principle of the model predictive control design
methods presented in this book.
After introducing the background information, in Chapter 6, we begin with
the topics in continuous-time model predictive control (CMPC). It is natural
that when the design model is embedded with integrators, the derivative of the
control signal should be modelled by the orthonormal basis functions, not the
control signal itself. With this as a start point, systematically, we will cover the
principles of continuous-time predictive control design, and the solutions of
the optimal control problem. It shows that when constraints are not involved
in the design, the continuous-time model predictive control scheme becomes
a state feedback control system, with the gain being chosen from minimizing
a finite prediction horizon cost. The continuous-time Laguerre functions and
Kautz functions discussed in Chapter 5 are utilized in the design of continuous-
time model predictive control.
In Chapter 7, we introduce continuous-time model predictive control with
constraints. Similar to the discrete-time case, we will first formulate the con-
straints for the continuous-time predictive control system, and present the
numerical solution of the constrained control problem using a quadratic pro-
gramming procedure. Because of the nature of the continuous-time formu-
lation such as fast sampling, there might be computational delays when a
quadratic programming procedure is used in the solution of the real-time op-
评论0