摘要
变压器故障诊断正确率取决于诊断模型的构建、特征参量的选取以及故障类型数据的丰富,模型参数的调整也变得尤为重要.针对依据经验调整参数导致随机森林模型诊断变压器故障正确率不够高的问题,提出了基于粒子群优化随机森林(PSO-RF)的故障诊断方法.以油中溶解气体的无编码比值构成特征参量作为模型输入,通过PSO算法搜索RF模型的两个最优参数(子树棵数和分裂特征数),建立PSO-RF模型诊断故障类别,并与不同特征参量选择方法和不同模型进行对比分析.运用两个具体实例的诊断结果来验证所提诊断模型和特征选取的有效性.结果表明:以无编码比值作为特征参量能挖掘更多的故障信息,并且PSO-RF模型故障诊断正确率优于SVM、BPNN与RF模型,随着样本空间的增大,故障诊断模型的效果越好.
The accuracy of transformer fault diagnosis relies on diagnostic model construction,feature parameter selection and the acquired fault type quantities,and model parameter adjustment is particularly important.Adjusting parameters by experience results in lower accuracy of the random forest model based transformer fault diagnosis.To improve the accuracy,this paper proposes a fault diagnosis method based on particle swarm optimization and random forest.The method forms the feature vector as the model input with non-code ratios of dissolved gases in the oil and then gives a PSO algorithm to search for two optimal parameters(the number of trees and the number of splitting features)of the RF model.Once the PSO-RF model is established,it will diagnose the fault type and compare the different feature parameter selection methods and different models.The diagnosis results of two specific examples are given to show the effectiveness of the proposed diagnostic model and feature selection.It can be seen that the feature selection based on the non-code ratios gets more fault information and the fault diagnosis accuracy of PSO-RF model is better than those from SVM,BPNN and RF models.It also predicts that if the sample space is increased,this proposed fault diagnosis model will give even better solution.
作者
李鹤健
徐肖伟
王科
赵勇军
吴世浙
刘可真
LI Hejian;XU Xiaowei;WANG Ke;ZHAO Yongjun;WU Shizhe;LIU Kezhen(Dali Power Supply Bureau,Yunnan Power Grid Co.Ltd.,Dali,Yunnan 671000,China;Electric Power Research Institute,Yunnan Power Grid Co.Ltd.,Kunming 650217,China;Yunnan Electric Power Technology Co.Ltd.,Kunming 650217,China;Faculty of Electric Power Engineering,Kunming University of Science and Technology,Kunming 650500,China)
出处
《昆明理工大学学报（自然科学版）》
北大核心
2021年第3期94-101,共8页
Journal of Kunming University of Science and Technology(Natural Science)
基金
国家自然科学基金项目(51477100)
云南电网有限责任公司科技项目(YNKJXM20180736)。
关键词
变压器
故障诊断
随机森林
粒子群优化
无编码比值
transformer
fault diagnosis
random forest
particle swarm optimization
non-code ratios
