Recurrent Network and Recursive Estimation Methods for State Space Modeling

Abstract

State space mpdeling for npnlinear and time-varying systems using hybrid intelligent methods is presnted A recurrent neural network and several recursive parameter estimation methods are utilized in this study.Convergence property of the recurrent network is first evaluated.Then the proposed method is implemented to identify a nonlinear system.Recursive least-sqursive least-squares with exponential forgetting, stochastic approximation,and projection algorithm are also given for comparison. Hybrid parameter estimation combined with evolutionary computation is then proposed . Simulation results show that the proposed schemes have better performance on system modeling then the conventional least-squares estimation and the recurrent neural network.