Integral minimum variance control and benchmarking

Abstract

A minimum-variance control problem is defined that involves the minimisation of the integral of estimation error. This introduces integral action into the controller. It is believed that the resulting solution is more valuable for closed-loop performance assessment and benchmarking, than the usual minimum-variance results. This is because most industrial controllers need to have integral action and hence the proposed benchmark, which is based on a controller that includes an integrator, will often be more appropriate. The advantage over many other benchmarking methods lies in the simplicity of the results. The situation where the controller structure has a limited, or restricted structure, is also considered. That is, a method is presented where the controller structure may be prespecified, and the coefficients obtained by direct parameter optimisation. This provides a more direct basis for comparison with controllers implemented in existing plants, that may only include, say PID, restricted structures.