Designing predictors for MIMO switching supervisory control (Q2719853)

A fundamental problem of switching supervisory control, that is choosing amongst a finite number of candidate controllers one which yields the best behaviour for the loop made up by the feedback interconnection with an uncertain and possibly time-varying MIMO plant, is under investigation. It is shown that the measure of mismatch between the loop made up by the plant in feedback with the candidate controller and the nominal ``tuned-loop'' related to the same candidate controller can be computed in real time by filtering an output prediction error without directly checking the behaviour of all possible loops via their implementation. The result is based on the introduced concepts of virtual reference (VR) and VR-normalized discrepancy (VRND). It is also shown that in the VRND-based switching supervisory control the noise becomes cooperative with the switching task. A benchmark problem (the positioning of serially connected four carts) is used as a running example to clarify the concepts and design choices, as well as to demonstrate the significant performance improvements achievable by the VRND filtering.