基于HHT和概率神经网络的变压器局部放电故障识别

摘要

为解决传统傅里叶变换和小波分解对变压器局部放电信号非平稳性的分析缺陷,以及BP神经网络易陷入局部极小点等问题,提出一种基于希尔伯特能量聚类和概率神经网络的变压器局部放电识别算法.算法利用希尔伯特?黄变换提取局部放电信号的希尔伯特能量谱,然后进行指数族聚类计算获得特征值,最后利用概率神经网络进行分类识别.分别对油中悬浮放电、沿面放电等放电类型进行模拟实验,并用此算法进行分析,实验结果表明,该算法所提取的特征值有较高的可分性,且分类识别率高,可以有效地识别变压器局部放电故障类型.%Since the traditional Fourier transform and wavelet decomposition have the defect for the non?stationary analysis of transformer partial discharge(PD)signal,and the BP neural network is easily to fall into the local minimum,a transformer partial discharge identification algorithm based on Hilbert energy clustering and probabilistic neural network(PNN)is proposed. The Hilbert?Huang transform(HHT)is used to extract the Hilbert energy spectrum of PD signals,and then the exponential family calculation is performed to obtain the feature values. The PNN is used to classify and identify the feature values. The simulation experiment was carried out for the discharge types of suspended discharge and surface discharge in oil,which are analyzed with the proposed algorithm. The experimental results show that the feature values extracted by this algorithm has high separability, and the algorithm has high classification identification efficiency,and can identify the fault types of transformer PD effectively.