Extended Kalman Filter Based Modified Elman-Jordan Neural Network for Control and Identification of Nonlinear Systems

摘要

In this paper, the Extended Kalman Filter (EKF) is used for online training of a recurrent neural network (RNN) model since the EKF outperforms the first order gradient-based algorithms as a second order method. The modified Elman-Jordan Neural Network model with one hidden layer is adopted as the RNN structure. Self-connections are added in context units to investigate their effects. Then, this model is utilized for identification and online control of a nonlinear single input single output (SISO) process model. The performance of the proposed structure is evaluated by simulation results. The effects of some parameters and the number of hidden units to the performance are also examined.