Identification of nonlinear time varying systems using recurrent neural networks

In this paper the identification using recurrent neural networks based on extended Kalman filter is presented. As it is well known, it is difficult to identify a nonlinear time varying system using traditional identification approaches. Although there have been some network architectures and learning algorithms for the nonlinear time variant systems, the lagged orders must be estimated. There is no need for a priori knowledge for the lagged orders in the recurrent networks. In this paper the learning algorithm has the fast convergence of the extended Kalman filter and needs no estimate of the lag in the system in the presented recurrent networks. Simulation results demonstrate the effectiveness and the fast convergence and good tracking capability of this approach.