Wavelet Neural Network Algorithm Based on Adaptive Unscented Kalman Filter and Its Applications
Xusheng Gan · Fire Control and Command Control · 2010
Wavelet Neural Network(WNN) trained by Extended Kalman Filter(EKF)has the shortcomings of slow convergent rate,no high precision and computing the Jacobian matrix etc.In the paper,on the basis of adaptive Kalman filter theory,an WNN training algorithm which is based on Adaptive Unscented Kalman Filter(AUKF) is proposed.The algorithm introduces into an adaptive factor in the frame of Unscented Kalman Filter(UKF) which is used to adjust the proportion of observation covariance and state parameter covariance of UKF,and to make the covariance of the predicated vector approach the true value.This can efficiently improve the precision of WNN.The simulation result show that WNN based on AUKF has the fast convergent rate,high estimation precision without computing the Jacobian matrix,and is fit for the problem of modeling and prediction for the nonlinear system.