基于小波网络的信号识别算法

Research of Time Series Based on Wavelet Neural Network

为提高小波网络运行速度,缩短小波网络的训练及运行时间,提出一种基于提升小波变换和神经网络算法的改进小波网络——提升小波网络。首先将带有明显特征的信号作为网络输入,经过权值处理输入到隐层节点进行提升小波变换处理,提取信号的低频系数作为隐层节点的输出,再经过权值化处理输入到输出层节点进行0-1输出,进而达到对信号的特征识别。该算法不仅继承了小波网络的多分辨分析、并行分布式处理、高自由度、强大的函数逼近能力等特性,同时具有提升小波变换的计算量小、本位运算速度快等优点。仿真实验结果显示,该算法在同等条件下更容易实现,并能保证运行精度,同时缩短训练和实际运行时间,对小波网络的工程应用具有一定指导意义。

For the improvement of speed,along with the reduction of training and running time of Wavelet Network,a Lifting- wavelet Neural Network based on the Lifting-wavelet transform and Artificial Neural Network is proposed.Time series with character is used as input signal,and processed by lifting-wavelet transform in the hidden layer after weight processing. Then the low frequency coefficient is picked up as the output signal of hidden layer,which would also be used as input of output layer after weight processing.Eventually the character identification of time series is approached by 0-1 output of output layer.The algorithm proposed in this paper not only inherited the multi-resolution analysis,distributed parallel pro- cessing,high degrees of freedom,and powerful function approximation ability from traditional wavelet neural network,but also introduced the strongpoint of less computation,faster speed and so on.