基于密度峰值聚类算法的局部放电脉冲分割

Partial Discharge Pulse Segmentation Based on Clustering by Fast Search and Find of Density Peaks

局部放电(PD)信号处理是电力设备绝缘状态评估的基础,而特征量提取又是信号处理的关键环节。特征量提取包括局部放电脉冲分割和放电特征量提取两个步骤。放电脉冲分割提取是后续PD信号特征提取及故障分类的前提。为尽可能保留放电信息,同时减少人工干涉,该文提出了一种基于聚类算法的PD脉冲分割提取方法。该方法采用小波分解算法进行滤波处理,使用噪声抑制比(NRR)表征滤波效果;以所有局部放电信号半波脉冲为对象,计算各半波脉冲的能量(即信号瞬时值平方对时间的积分),从而使该方法能更准确地描述局部放电过程。应用Otsu算法自适应计算能量阈值并结合密度峰值聚类算法(DPC)实现PD脉冲的自动分割。在实验室建立了三种不同类型局部放电模型,采集得到10组电晕放电、11组悬浮放电和30组锥板放电数据,以对该文方法进行验证。结果都取得了80%以上的识别率,比同类算法更高或相当,表明了该文方法的优越性。

Partial discharge(PD)signal processing is the basis of insulation state evaluation of electric power equipment,and feature extraction is the key link of signal processing.The feature extraction includes two steps:partial discharge pulse segmentation and discharge feature extraction.Pulse segmentation extraction is the premise of subsequent feature extraction and fault classification of PD signals.In order to preserve discharge information as much as possible and reduce manual interference,a PD pulse segmentation method based on clustering algorithm is proposed in this paper.The wavelet decomposition algorithm is used for filtering,and the noise rejection ratio(NRR)is used to characterize the filtering effect.The method takes all the partial discharge half-wave pulses into consideration to calculate the energy of each half-wave which is the integral of squared instant value of the PD signal over time.Therefore,it can describe the process of partial discharge more accurately.The Otsu algorithm is used to calculate the energy threshold adaptively,and one density peak clustering algorithm,called as clustering by fast search and find of density peaks(DPC),is used to achieve automatic segmentation of PD pulses.Three types of partial discharge models are established in the laboratory.Data from 10 sets of corona discharge,11 sets of suspension discharge and 30 sets of cone-plate discharge are collected to verify the method.The results have achieved a recognition rate of more than 80%,which is higher or equivalent to the existing similar algorithms,demonstrating the superiority of this method.