Model structure selection for switched NARX system identification: A randomized approach

The identification of switched systems is a challenging problem, which entails both combinatorial (sample-mode assignment) and continuous (parameter estimation) features. A general framework for this problem has been recently developed, alternates between parameter estimation sample-mode assignment, solving tasks to global optimality under mild conditions. This article extends the nonlinear case, further aggravates complexity since model structure selection task be addressed each mode system. To solve issue, we reformulate learning in terms optimization probability distribution over space all possible structures. Then, randomized approach employed tune distribution. performance proposed on some benchmark examples analyzed detail.