In this talk, we consider a unified probabilistic approach to model estimation/selection and controller design, and we deal with the complexity of the model and the controller. The objective systems are assumed to include unknown random parameters with probability distributions. The first issue is what evaluation function, for model estimation, is reasonable with respect to the controller design. Thus, we show that the complexity of the distributions of the unknown system parameters and the control performance should satisfy some conditions for the unified model estimation/selection and controller design to be well-posed. Moreover, we analyse the effects of the complexity of the parameter distribution model and the class of controller on the expectation of the evaluation functions for model estimation, where the expectation can be used as the criterion for model selection and for the selection of the class of controller.
Back to Control Seminars Page