提出一种用于非线性函数逼近的小波神经网络算法 ,分析了网络的拓扑结构 ,给出了网络的参数估计方法 .采用遗忘因子法训练网络的权值 ,利用具有优良渐近性质的递推预报误差算法训练尺度因子和平移因子 ,分析并给出两种小波元的个数选择方法 .该算法用于非线性函数逼近时优于同等规模的 BP神经网络 .仿真研究表明 ,该方法具有收敛速度快 ,逼近精度高等优点 ,在为非线性系统建模提供一种新方法的同时 ,也为复杂非线性系统的辨识提供有益的参考 A wavelet neural network algorithm for approximating non-linear functions is proposed. The topology of the network is analyzed, and the parameter estimation method of the network is given. The weight of the network is trained by the forgetting factor method. Recursion with good asymptotic properties The prediction error algorithm training scale factor and the translation factor are analyzed and two number selection methods are given.The algorithm is better than the same scale BP neural network when the nonlinear function is approximated.The simulation results show that this method has the advantages of convergence Speed, high approximation accuracy, provide a new method for nonlinear system modeling, but also provide a useful reference for the identification of complex nonlinear systems