基于多小波包系数熵和人工神经网络的输电线路故障类型识别方法

Transmission Lines Fault Recognition Method Based on Multi-wavelet Packet Coefficient Entropy and Artificial Neural Network

基于多小波包分解系数和信息熵的概念定义了多小波包系数熵的表达式,并提出多小波包系数熵和人工神经网络相结合的输电线路故障类型识别方法:首先对不同故障工况下采集的故障电流信号进行适当的多小波包分解,计算各频带的系数熵;然后构造多小波包特征向量,将这些向量作为训练样本对径向基函数(radial basis function,RBF)神经网络进行训练;当输电线路发生故障时,将提取的故障电流信号的多小波包系数熵特征向量输入训练好的RBF神经网络,即可实现故障类型的识别。仿真结果表明采用多小波包提取的故障电流特征量比采用传统小波包提取的特征量信息更丰富,对人工神经网络的训练效果更好,网络识别精度具有明显优势。

Based on the decomposition coefficient of multi-wavelet packet and the concept of information entropy the expression of multi-wavelet packet coefficient entropy （MPCE） is defined, and the method to recognize fault types of transmission lines by combining multi-wavelet packet coefficient entropy with artificial neural network （ANN） is proposed. First, the fault current signals sampled under various fault conditions are decomposed by multi-wavelet packet properly and the coefficient entropies of different frequency bands are calculated; then the eigenvectors of multi-wavelet packet are constructed, and taking these eigenvectors as training samples the radial basis function （RBF） neural network is trained; when fault occurs in transmission line, the fault type recognition can be implemented by inputting the extracted MPCE eigenvector of fault current signal into the trained RBF neural network. Simulation results show that there is more fault current characteristic information extracted by multi-wavelet packet than that extracted by traditional wavelet packet and a better training result of ANN by MPCE eigenvectors can be obtained, meanwhile, the recognition precision of network is better than that by traditional wavelet packet.