Adaptive control of plants subject to changes of structure and parameters using a multi-model approach

In this paper, the outline of an adaptive controller based on a multi-model approach is presented. This control concept is applicable as well for systems subjected to sudden changes as for nonlinear plants. The multi-model consists of several so called sub-models which describe the characteristics of the system at a certain operating-point. Here, the individual sub-models are linear ARMAX-models of different dynamics and structure. Since different sub-models describing the various operating-points of the system can be determined successively, the multimodel approach can be attached to the group of the learning methods. Several algorithms to detect sudden changes of the plant and to select the best sub-model which meets the actual situation are described. Moreover, an online estimation of the actual model is necessary in order to identify new sub-models, in case additional operating-points occur. Based on the actual model of the system, the manipulated signal is computed using an indirect adaptive control algorithm, at every sampling instant.