基于灰度图像分解的局部放电特征提取与优化  

Partial discharge feature extraction and optimization based on gray image decomposition

作　　者：汪可[1] 张书琦[1] 李金忠[1] 孙建涛[1] 赵晓宇[1] 廖瑞金[2] 邹国平 WANG Ke;ZHANG Shu-qi;LI Jin-zhong;SUN Jian-tao;ZHAO Xiao-yu;LIAO Rui-jin;ZOU Guo-ping(China Electric Power Research Institute,Beijing 100192,China;State Key Laboratory of Power Transmission Equipment&System Security and New Technology,Chongqing University,Chongqing 400044,China;State Grid Zhejiang Electric Power Research Institute,Hangzhou 310014,China)

机构地区：[1]中国电力科学研究院,北京100192 [2]重庆大学输配电装备及系统安全与新技术国家重点实验室,重庆400044 [3]国网浙江省电力公司电力科学研究院,浙江杭州310014

出　　处：《电机与控制学报》2018年第5期25-34,共10页Electric Machines and Control

基　　金：国家电网公司科技项目(5211DS16000G)

摘　　要：提取有效的局部放电(PD)特征是输变电设备缺陷识别的前提。以局部放电灰度图像为分析对象,提出了基于二维主成分分析(2DPCA)的局部放电图像特征提取策略。算法通过2DPCA将PD灰度图像分解为多个一维向量,并对每个向量提取了9个特征参量,组成了PD图像分解特征集。同时,建立了基于粒子群优化(PSO)算法的PD特征选择算法,以优化PD图像分解特征,提升局部放电缺陷类型识别结果。对实验室考虑多因素影响的PD样本识别结果表明,2DPCA图像分解特征可以取得93%的PD缺陷识别率,经过PSO优化后的2DPCA特征可以将PD识别率提高至96%,并且特征维数由72降至28,充分说明方法的有效性。另外,对添加不同随机干扰的PD样本平均识别率均大于85%,表明2DPCA图像特征具有较好的抗随机干扰能力。Effective features extraction of partial discharge(PD)is the foundation of defect identification of electrical apparatus.Using PD gray image as the analysis object,a PD image features extraction strategy was proposed based on two-dimensional principal component analysis(2DPCA).Various 1-dimensional(1D)vectors were obtained by implementing 2DPCA on PD gray images in the proposed method.9 characteristic parameters were extracted from each 1D vector,which constituted the PD image decomposition features.In addition,a PD features selection algorithm was developed based on particle swarm optimization(PSO)algorithm,which attempts to optimize the extracted PD image decomposition features and improve the PD recognition accuracy.The recognition results of PD samples considering the multi-factor influences in laboratory illustrate that the proposed 2DPCA image decomposition features can achieve the high PD recognition accuracy of 93%.Besides,the PSO optimized 2DPCA features can further improve the PD recognition accuracy to 96%and simultaneously reduce the feature dimension from 72 to 28,which fully demonstrates effectiveness of the proposed algorithm.Moreover,the average recognition accuracies of PD samples added with different random noises are all higher than 85%,which indicates that 2DPCA image features possess good tolerance ability of random noises.

关 键 词：局部放电 模式识别 图像分解 特征提取 特征选择 模糊k近邻 

分 类 号：TM835[电气工程—高电压与绝缘技术]