One-Step Weights Updating Back-Propagation Neural Network for Nonlinear System Identification

摘要

Given that many filters such as least mean squares and recursive least squares are not able to deal with nonlinear system. In this paper, a nonlinear system identification technique using a specially designed neural network is investigated. Precisely, a power-activated back-propagation neural network is first constructed. Then, a high efficient weights updating method which only requires one-step iteration in its training session is presented. The system identification performance is evaluated through MATLAB simulations. The simulation results validate the one-step weights updating method and show satisfactory nonlinear system identification performance.