A Novel Criterion for Optimal Identification of Wiener Models Using Local Linear Models

A novel criterion is presented for the optimal choice of local linear models representing a Wiener model. The linear part is modeled by a transfer function and the nonlinear (possibly not invertible or discontinuous) function is modeled by a series of local linear models. The algorithm is based on a recursive approach where the novel criterion is applied to test the local models for adhering a predefined accuracy. The criterion automatically adapts to existing output noise levels and delivers an optimal model with respect to a minimum number of local models and guaranteed global performance. Furthermore, the confidence and prediction intervals of the identified model are supplied. Simulation results demonstrate the performance of the proposed criterion.