一种基于BP算法学习的小波神经网络

A Kind of BP Algorithm-Learning Wavelet Neural Network

为发展 Szu的基于信号表示的小波神经网络 ,提出一种多输入多输出的小波网络模型 ,网络隐层采用框架小波函数、输出层采用 Sigmoid激励函数 ,并选用“熵误差函数”以加速网络的学习速度。奇偶判别和混沌时间序列预测例子的实验结果表明了它具有良好的函数逼近能力和推广能力 ,收敛速度和均方误差均优于相同结构的多层感知器模型。

This paper develops the Szu′s signal representation based wavelet neural network (WNN) and proposes a kind of multi input multi output WNN model. The sigmoid function and wavelet basis function satisfying the frame condition are employed as an activation function in the output and hidden layer respectively, and the entropy error function is also used to accelerate the learning speed. The experimental results on parity problem and chaotic time series prediction demonstrated that this WNN has excellent functional approximation and generalization abilities, and the convergence speed and the mean square error (MSE) also show its superiority to a multilayer perceptron (MLP) model with the same architecture.