基于遗传算法的电网过电压分层模糊聚类识别  

Hierarchical Fuzzy-clustering Classification of Overvoltages in Power Systems Based on the Genetic Algorithm

作　　者：杜林[1] 郭良峰[1] 司马文霞[1] 陈明英[1] 赵立进 

机构地区：[1]输配电装备及系统安全与新技术国家重点实验室(重庆大学),重庆市沙坪坝区400044 [2]贵州电力试验研究院,贵州省贵阳市550002

出　　处：《中国电机工程学报》2010年第10期119-124,共6页Proceedings of the CSEE

基　　金：国家重点基础研究发展计划项目(973项目)(2009CB-724504)~~

摘　　要：过电压识别对过电压起因及故障分析,改进输电线路和变电站设备绝缘配合具有重要意义。提出了基于小波多分辨率能量分布的电力系统过电压特征参量提取方法,针对特征向量存在交叉重叠的情况,引入分层模糊聚类识别的方法。构建分层识别的过电压分类树,通过对各种特征向量进行归纳分析与综合。特征向量集按不同的模块层次选取,形成模块层次结构,构成该层最佳特征量。将遗传算法的全局搜索和并行特性引入到模糊聚类中,弥补了模糊C-均值聚类(fuzzyC-means,FCM)算法存在局部性搜索和对初始聚类中心敏感等不足,通过全局搜索与局部搜索相结合的方式提高收敛速度,并加入移民策略来维持群体多样性,将该方法应用于实际过电压数据模式识别分类中,结果表明该方法能有效降低误分类率,从而对电力系统过电压类型进行有效识别。Overvoltage recognition is of great importance for the analysis of origins of overvoltages and the induced failures, the improvement of insulation coordination between the power transmission lines and the electrical apparatus. A method to obtain the characteristic quantities of overvoltages is proposed based on the energy distribution of wavelet transform. In view of the eigenvectors overlap, a hierarchical classification method for overvoltage recognition was applied . By constructing a multi-layer overvoltage classification tree, the eigenvector set was selected from different modules and different levels, forming the module-level structure, and constituting the best features of the layer. By introducing the genetic algorithm into fuzzy clustering, the problems were solved about the locality and sensitiveness of the initial condition in fuzzy C-means （FCM） clustering and the convergence rate was improved by combining the global search with the local search and employing the migration strategy. The method was applied to the pattern recognition classification for field overvoltages of the power system, and results show that the method can reduce the error rate of classification greatly, and is effective to recognize overvoltages.

关 键 词：过电压 分层识别 分类树 遗传算法 模糊聚类 

分 类 号：TM86[电气工程—高电压与绝缘技术]