基于DGA的差分进化支持向量机电力变压器故障诊断  

Fault Diagnosis of Power Transformer Based on DGA of Algorithm for SVM

作　　者：贾立敬 张建文[1] 王传林[1] 严家明[1] 周贤娇[1] 陈鲁娜 

机构地区：[1]中国矿业大学信息与电气工程学院,江苏徐州221116

出　　处：《高压电器》2015年第4期13-18,共6页High Voltage Apparatus

基　　金：第54批博士后基金(2013M541755)~~

摘　　要：针对支持向量机中参数选择严重影响分类效果的特点,提出采用差分进化算法对核函数g和惩罚因子c进行优化,得到最优的支持向量机模型,用于变压器的故障诊断。在简单介绍支持向量机的基础上,分析了采用差分进化算法对支持向量机优化的可行性。通过将收集的数据样本进行预处理,再利用差分进化的变异、交叉和选择对高斯径向基核函数进行优化,搜索出最优(c,g),并对得到的参数进行验证,获得最优的支持向量机模型。仿真实验表明,与SVM、GRID-SVM、GA-SVM、PSO-SVM相比,该方法误判率最低、全局寻优能力及鲁棒性较好。In response to parameter selection seriously affect classification results in SVM, this paper proposes a method using differential evolution optimizes on parameter of kernel function g and penalty parameter c, aiming to getting the SVM machine model for the fault diagnosis of transformer. Firstly, this paper analyzes the feasibility of using differential evolution algorithm for SVM optimization based on the brief introduction of SVM. Through pretreating the collecting data samples and using the mutation,crossover and selection of differential evolution optimizes on RBF, the best parameters c and g are found out, and the parameters to obtain the SVM model are verified. The results show that, compared with non-optimized SVM, GRID-SVM, GA-SVM and PSO-SVM, this method has the lowest misjudgment rate and good global search ability and better robustness.

关 键 词：支持向量机 差分进化 核函数 故障诊断 

分 类 号：TM41[电气工程—电器]