On the determination of an on-demand policy for a multilayer control system

The cost-performance tradeoff problem associated with a multilayer control system for controlling a class of static, nonlinear, multivariable systems is considered. The multilayer control system has a number of layers of control functions each of which updates different subsets of the control variables at different costs. A favorable cost-performance tradeoff is achieved by determining at each control decision time which subset of the control variables is to be updated. In this paper, we present a mathematical model which describes the operation of the multilayer control system. Also we show that the problem of determining a decision rule (policy) which results in an optimal cost-performance tradeoff can be formulated as a problem in Markovian decision processes. Consequently, an optimal policy can be identified by solving a linear program. In order to reduce the computational effort required for identifying the optimal policy, a class of parameterized policies is introduced based on a measure of deviation of the disturbance. This approach provides a designer with a practical method of determining a control policy which achieves a favorable cost-performance tradeoff. An example is given for demonstrating a possible application to process control.