A new fast U-D factorization-based learning algorithm with applications to nonlinear system modeling and identification

Abstract

A new fast learning algorithm for training multilayer feedforward neural networks by using variable forgetting factor technique and U-D factorization-based fading memory extended Kalman filter is proposed. In comparison with the backpropagation (BP) and extended Kalman filter (EKF) based learning algorithms, the proposed algorithm can provide much more accurate learning results in fewer iterations with fewer hidden nodes as well as improve convergence rate and numerical stability. In, addition, it is less sensitive to the choice of initial weights and initial covariance matrix as well as other setup parameters. Simulation results of nonlinear dynamic system modeling and identification show that the new algorithm proposed here is an effective and efficient learning algorithm for feedforward neural networks.