以小波理论为基础，讨论了小波径基函数网络非线性辨识的基本原理，并提出从全部小波基中选出对辨识最起作用的一部分小波基的正交优化方法，以解决小波神经网络隐层神经元数量过多的问题。用推广 Ｋａｌｍ ａｎ 滤波方法训练小波网络，大大加快了它的收敛速度。仿真结果表明了该方法的可行性和有效性。 On the basis of wavelet theory, the basic principle of nonlinear identification of wavelet radial basis function network is discussed. An orthogonal optimization method is proposed to select a part of wavelet bases which are most effective for identification from all wavelet bases, so as to solve the problem of wavelet neural network hidden Layer neurons in the number of problems. Using the Kalman filtering method to train the wavelet network greatly accelerates its convergence speed. Simulation results show that the method is feasible and effective.