Multiple Model Adaptive Control Using Dual Youla-Kucera Factorisation

We propose a multi-model adaptive control scheme for uncertain linear plants based on the concept of model unfalsification. The approach relies on examining the ability of a pre-computed set of plant-controller candidates and choosing the one that is best able to reproduce observed in- and output signal samples. The ability to reproduce observations is measured as an easily computable signal norm. Compared to other related approaches, our procedure is designed to be able to handle significant measurement noise and closed-loop correlations between output measurements and control signals.