Back to publications...
Control of plants with changing dynamics using switching model predictive control
Simon Lam and Edward J. Davison
Abstract
A recent problem in adaptive control is the task of controlling a plant which is subject to large structural
changes. In this case, it is often assumed the plant can be described by a specified family of models, and that
the plant can change from one member of the family of models to some other member of the family of models. Such a
system can be controlled by using a set of predefined controllers, and then switching among the controllers to
compensate for the change in the plant dynamics. In previous studies, such switching controllers have been found
to compensate for changes in the plant dynamics, but the transient response error is typically excessively large,
due to the fact that the switching controller, during its "learning period", may produce unstable closed-loop
systems, which produce excessively large control signals. In this paper, a new switching controller is described,
which uses model predictive control to impose constraints on the control signals. It is shown by simulation that,
since the new controller does not produce "peaking" in the control signals, it subsequently solves the above
family of controllers switching problem with greatly improved transient behaviour.
