「NEW RESULTS ON MODEL PREDICTIVE CONTROL (MPC)」

東京電機大学客員教授, トロント大学名誉教授　Edward J. Davison

計測自動制御学会 制御理論部会, 東京電機大学 21 世紀 COE 研究推進室 共催

Abstract

One of the most attractive features of Model Predictive Control (MPC) is its ability to deal directly with constraints, and as a consequence, it is often applied in the control of industrial chemical factories and in other such industrial systems.

There are however often some significant disadvantages of MPC, such asthe excessive CPU time requirements which are often required in order to compute the optimal control for the system, which has the effectthat it limits the application of MPC to those processes which have slow dynamics such as which occur in process control. Other concerns about MPC are that the resulting transient response obtained may have undesirable cross- channel interaction effects, and undesirable transient behaviour.

This talk describes a new approach to MPC which addresses the above problems. In particular, the talk describes a new algorithm for solving MPC, based on applying a new proposed non-feasible active set method, to compute the optimal control which minimizes a new proposed MPC performance index, and which have the notable features that:

• The CPU computing time required to solve the MPC problem is significantly faster than conventional MPC algorithms, (e.g. it may be some 40x faster than standard MPC algorithms).
• The resulting optimal control obtained has the property that it produces a smooth, non-oscillatory output response, which has small cross-channel interaction effects.
• The resulting control obtained achieves tracking and regulation against un-measurable disturbances, not only for constant tracking and disturbance signals, but also for the class of unbounded signals called “extended constant signals”. Examples of such signals include the class of unbounded signals given by w[k] =sqrt(k), k = 0,1,2,3,….