BP小波神经网络快速学习算法研究

Fast Training Algorithm of BP Wavelet Neural Network

讨论了BP小波神经网络在训练过程中减小误差函数时最优方向的确定和自适应调整学习率的方法。首先论证了小波神经网络的数学基础 ,然后讨论了BP小波神经网络的学习过程 ,重点讨论了减小误差函数最优方向的确定方法 ,即如何保证步长方向与负梯度方向一致 ,由此得出了自适应调整学习率的简便方法。该方法具有普遍性 ,有广泛的应用价值。仿真结果表明 ,采用最优梯度下降方向可以大幅度提高BP小波神经网络的学习速度。

The determination of the best direction of step length vector and the adaptive adjustment of training rate for BP wavelet neural network (WNN) are discussed. Firstly, the mathematical foundation of WNN is presented, then the training process of WNN, esp. the determination method of best direction in reducing error function as well as the adaptive adjustment of training rate is discussed. Some new conclusions are put forward. The simulation result shows the training speed of WNN can be improved greatly. The method is general and can be applied extensively.