论文部分内容阅读
局部放电信号特征提取及模式识别是电力设备状态评估和故障诊断的关键,基于局部放电波形的时频分析能够全面获取波形特征,但目前常用的时频分析方法存在交叉项干扰问题,会产生“虚假信号”。为解决该问题,提出将集合经验模态分解(EEMD)和Cohen类分布结合的方法应用于局部放电信号的时频分析,并运用回归型支持向量机(SVR)抑制EEMD端点效应。利用指数衰减振荡函数加入Gaussian白噪声和窄带干扰来模拟电力设备高频电流局放信号。结果表明,该方法能够准确辨识出特征脉冲信号,且在保证信号项时频聚集性的同时,有效削弱信号内部交叉干扰项。利用该方法对变电站现场实测局放信号进行时频分析,所得结果验证了其在现场干扰环境下的有效性和实用性。
Partial Discharge Signal Feature Extraction and Pattern Recognition are the keys of power equipment condition assessment and fault diagnosis. Based on the time-frequency analysis of partial discharge waveform, the waveform characteristics can be fully obtained. However, the cross-term interference problem exists in the commonly used time-frequency analysis method, “False signal”. In order to solve this problem, a new method based on EEMD and Cohen distribution is proposed for the time-frequency analysis of PD signals. The regression model of support vector machine (SVR) is used to suppress the EEMD endpoint effect. Exponentially decaying oscillation function is used to add Gaussian white noise and narrow-band interference to simulate high-frequency current signals of power equipment. The results show that this method can accurately identify the characteristic pulse signal, and effectively weaken the signal cross-interference terms while ensuring the signal time-frequency aggregation. The method is applied to time-frequency analysis of PD signals measured in substations, and the results verify the effectiveness and practicability of the method in the scene of interference.