On-Line Model Selection Techniques By Using Multiple Models And Supervision Algorithms

Abstract

In this paper, an intelligent adaptive multi-model based control scheme is proposed to obtain the lowest-order model for a system under control by means of adaptation and switching between multiple models. The multi-model scheme is composed of three models of increasing consecutive orders operating in parallel along with a switching mechanism between them. The switching policy selects on-line the necessary order of the nominal model required to achieve the desired degree of performance of the closed-loop depending on the reference signal selection. In this way, the order selection is performed automatically in real-time by comparing the actual behaviour of the system with the desired performance for the closed-loop. The estimation of the parameters of the model is performed by an adaptive algorithm integrating a so-called multi-estimation scheme. This architecture leads to a simple procedure to on-line select the appropriate order of the nominal model required for a certain control application with assessment of a prescribed level of closed-loop performance.