Application of Second Order Diagonal Recurrent Neural Network in Nonlinear System Identification

Abstract

In this paper, a kind of second order diagonal recurrent neural network (SDRNN) identification method based on dynamic back propagation(DBP) algorithm with momentum term is proposed. This identification method overcomes the disadvantages such as slow convergent speed and trapping the local minimum. The SDRNN is similar as diagonal recurrent neural network(DRNN) in the structure, two tapped delays are used in the hidden neurons of DRNN, the simple structure of the DRNN is retained, the identification of a nonlinear system is realized with SDRNN. Serial-parallel identification architecture is applied in the modeling. Simulation results show that improved algorithm is effective with advantages the fast convergence, higher identification accuracy, higher adaptability and robustness in system identification. It is suitable for real-time identification of dynamic system.