基于VMD-WVD分布与堆栈稀疏自编码网络的局放类型识别

Pattern Recognition of Partial Discharge Based on VMD-WVD and SSAE

针对目前常用的浅层模式识别方法对高维大容量样本处理困难的问题,提出一种基于变分模态分解(variational mode decomposition,VMD)的Wigner Ville分布(Wigner-Ville distribution,WVD)和堆叠稀疏自编码网络(stockedsparse auto-encoder,SSAE)的局部放电(partial discharge,PD)信号的模式识别方法。首先,以VMD算法对PD信号进行分解,对所得各分量进行时频分析得到相应的WVD;然后,以PD信号的VMD-WVD分布为输入量,利用SSAE对样本集合进行训练,自主提取内在特征。此外,将SSAE与稀疏自编码器(stackedsparseauto-encoder,SAE)的输出特征进行比较,验证了SSAE网络特征提取能力的优越性;最后,用训练好的SSAE网络完成测试样本的局部放电类型的识别。同时,以基于反向传播(backpropagation,BP)神经网络和支持向量机(support vector machine,SVM)的识别结果与该结果进行比较。结果表明,所采用的识别方法具有更高的正确识别率。

In allusion to the problem of common pattern recognition method in shallow architecture being unable to handle with high dimensional and large samples, this paper proposed a kind of partial discharge (PD) pattern recognition method based on variational mode decomposition (VMD) and stocked sparse auto-encoder (SSAE). Firstly, a PD signal was decomposed into several components, and the WVD would be obtained based on the analysis on these components. Then, the VMD-WVD was regarded as the input vector to train SSAE in order to extract inner features. In addition, the output features based on SSAE and SAE were compared and the results of comparison verified the superiority of SSAE. Finally, PD types of test samples were recognized by trained SSAE. Compared with the recognition results based on BP and SVM, it was shown that the results based on SSAE were better.