Criterion for selection of model and controller design based on I/O data

We propose information criteria not only for model estimation and selection but also for selection of the design method of a controller. The criteria are derived by the same approach of AIC/TIC, however we consider the modeling of the distribution of the control performance substituting for that of system index. The criteria are composed of a log-likelihood function and a bias term similar to AIC/TIC, and especially the bias term depends on the controller. Moreover, when the controller is designed under some conditions, the criteria become stochastic variables and their expectations are further added to by the other bias terms, which are in proportion to the dimension of the control parameter. This shows that model estimation, selection and control design are influenced simultaneously by the complexity of model and that of controller.