变压器局部放电信号检测与类型识别

摘要

A transformer partial discharge detection system based on ultra⁃high frequency(UHF)method was designed to realize the signal detection and type recognition of transformer partial discharge. The partial discharge experiments were conduct⁃ed for the four typical transformer discharge models to obtain the corresponding envelope signal data of partial discharge. The da⁃ta is uploaded to the computer through Ethernet communication. The envelop signal data was denoised by improving the dual⁃tree complex wavelet transform. It is not difficult to find from the denoised single that the envelop signal shapes of the same par⁃tial discharge model are almost the same,and the envelop signal shapes of different partial discharge models are different. The transformer partial discharge types were recognized with BP neural network,by extraction of characteristic parameters of six en⁃velop signals,and in combination with the external loading voltage. When the training error δ is 0.002,the average correctness of transformer discharge type recognition can reach up to more than 98%.%为实现变压器局部放电信号检测和类型识别，设计基于超高频（UHF）法的变压器局部放电检测系统，针对4种典型的变压器放电模型进行了局部放电实验，获得相应的局部放电包络信号数据，并通过以太网通信将数据上传至电脑。利用提升双树复小波变换对包络信号数据进行消噪，从消噪后的信号不难看出，同一放电模型的局部放电包络信号形状大致相同，不同放电模型存在差别。提取6种包络信号的特征参数，结合外部加载电压，采用BP 神经网络对变压器局部放电类型进行识别，当训练误差δ=0.02时，变压器放电类型识别平均正确率在98%以上。