基于时频特征核熵成分分析的局部放电模式识别方法

Partial Discharge Pattern Recognition Method Based on Kernel Entropy Component Analysis of Time-frequency Features

为实现电气设备局部放电模式的准确识别,提出了一种基于时频特征核熵成分分析的局部放电模式识别方法。首先采用S变换理论对局部放电脉冲信号进行时频特征分析,针对S变换分析结果维数庞大但冗余信息较多而不便于模式识别的缺点,基于核熵成分分析方法对S变换结果进行压缩降维处理,得到了局部放电模式识别时频特征向量,同时结合随机森林分类器实现了局部放电类型的准确识别。搭建了尖端放电、沿面放电、气泡放电、悬浮放电等典型变压器绝缘缺陷模型并采集了局部放电信号,分别采用文中方法、PCA方法及KPCA方法进行了局放模式识别实验。实验结果表明,相比PCA方法及KPCA方法,文中方法局放模式识别结果准确率较高且耗时较短。

To realize accurate pattern recognition of partial discharge in electrical equipment,a partial discharge pattern recognition method based on the kernel entropy component analysis of time-frequency feature was proposed. In this method,the time-frequency analysis of partial discharge pulse signal was carried out using the S-transform theory. Considering difficult pattern recognition of the S-transform due to a large number of dimensions of analysis results and too much redundant information, the kernel entropy component analysis(KECA) was adopted to compress the S-transform results. The time-frequency feature vector was obtained to recognize partial discharge type accurately by means of the random forest classifier. Furthermore,four typical transformer insulation defect models,which can produce point discharge,surface discharge,bubble discharge and suspended discharge,were set up in the laboratory,and the partial discharge signals were collected. Partial discharge pattern recognition experiments were conducted with the proposed method,PCA method and KPCA method, respectively, and the results showed that the proposed method obtained higher pattern recognition accuracy with shorter time.