采用CSA优化LS-SVM的油浸式变压器故障诊断方法  

A fault diagnosis method for oil-immersed transformer based on LS-SVM optimized by CSA

作　　者：高振兴[1] 骆玉海 李海山 郭创新[1] 

机构地区：[1]浙江大学电气工程学院,浙江杭州310027 [2]烟建集团有限公司,山东烟台264000 [3]烟台市供电公司,山东烟台264001

出　　处：《能源工程》2011年第3期14-19,共6页Energy Engineering

基　　金：国家自然科学基金资助项目(50677062);教育部新世纪优秀人才支持计划(NCET-07-0745)

摘　　要：提出了一种基于耦合模拟退火(CSA)优化最小二乘支持向量机(LS-SVM)的变压器故障诊断方法,该方法以变压器油中五种溶解气体含量为故障特征量,利用故障样本集结合CSA算法、k-折交叉验证来训练并优化一对一组合分类的LS-SVM模型,并将其应用于故障诊断分类。样本测试集验证表明,该方法可避免改良三比值法中编码缺失、编码边界过于绝对的问题,获得的支持向量机模型诊断效果比BP、RBF神经网络更加稳定,准确度也高于一对多组合分类的LS-SVM模型,具有良好的实用性。A fault diagnosis method for oil-immersed transformer based on LS-SVM which is optimized by CSA is presented.This method takes five kinds of dissolved gas content as fault characteristics,and integrates fault samples with CSA algorithm and k-fold cross-validation to train and optimize the LS-SVM model based on one-against-one（1-a-1） classifying combination.The derived model is applied in fault diagnosis classification.Testing set show that the coding loss and too absolute coding boundary problems with the improved three ratio method can be avoided by the method,and the derived LS-SVM model gets more stable diagnostic effects than BP,RBF neural network while its accuracy is also higher than the LS-SVM with one-against-all（1-a-a） classifying combination.The proposed method has good practical value.

关 键 词：变压器 故障诊断 LS-SVM CSA 

分 类 号：TM411[电气工程—电器]