Nonlinear Identification Based on Diagonal Recurrent Neural Network and Particle Filter
Deng Xiaolong, Zhou Pingfang · 2009
Diagonal recurrent neural network (DRNN) is widely applied to nonlinear identification. In this paper, the extended Kalman filter and particle filter are firstly combined to train DRNN. Utilizing time windows, a method to evaluate the dynamical performance of DRNN is presented. Network weights of particles are optimized by the resampling algorithm. The high convergent speed and high training precision are obtained by the new algorithm. Simulation results of the nonlinear dynamical identification verify the validity of the new algorithm.