Abstract
In this paper, we analyse and discuss the efficiency of an expanded likelihood and a criterion for model estimation and selection problem in consideration of controller design by computer simulations. In the model estimation and selection, the uncertainty of system is transformed to that of control performance by a controller, and the models are estimated on the transformed space. According to the proposed method, the models are estimated by expanded maximum likelihood method on the transformed space, and the class of models is selected by the proposed criterion. The simulation results show that the criterion evaluates both the uncertainty of control performance and the discrepancies between the true system and estimated models. We also show that a designer of the controller can adjust weights of the two above-mentioned properties by selecting proper weights on the likelihood.
