Optimal experiment design for identification of ARX models with constrained output in non-Gaussian noise

Abstract

The identification of ARX models with constrained output variance in the presence of non-Gaussian distribution of measurements is proposed in this paper. In the presence of non-Gaussian noise, the Masreliez–Martin filter (robust Kalman filter) is the natural frame for identification of model parameters. For the purpose of increasing the practical value of the filter, a heuristic modification is performed. Also, an optimal input is obtained by a minimum variance controller with a Gaussian reference signal. A fundamental issue in experiment design is that the optimal input in general depends on system parameters to be estimated. In order to realize obtained optimal input, authors have proposed a two-stage adaptive procedure, where one iterates between parameter estimation, on the one side, and experiment design using the current parameter estimates, on the other. The practical behavior of new approach to optimal input design for robust identification of ARX models is shown by intensive simulations.