基于XGBoost算法的变压器油中溶解气体分析  

Dissolved Gas Analysis in Transformer Based on XGBoost Algorithm

作　　者：沈晓峰 孙进 顾华 吴继健 张佳栋 SHEN Xiaofeng;SUN Jin;GU Hua;WU Jijiang;ZHANG Jiadong(State Grid Qingpu Supply Company of Shanghai Electric Power Company,Shanghai 201799,China)

机构地区：[1]国网上海市电力公司青浦供电公司,201799

出　　处：《电力大数据》2022年第9期20-29,共10页Power Systems and Big Data

摘　　要：针对目前常用的变压器油中溶解气体分析方法的不足,本文尝试根据油色谱在线监测系统产生的大量数据,利用机器学习算法对变压器故障进行诊断和鉴别。由于变压器故障诊断中存在样本小的特点,普通机器学习算法泛化能力较差,本文提出基于极端梯度提升(XGBoost)算法,结合油中气体和故障特征的理论基础,筛选气体并构造特征,基于特征重要度、网格搜索和交叉验证调参优化模型,使用SoftMax函数计算潜在故障预警。以334组变压器的油中溶解气体含量作为算例进行验证,XGBoost算法可以对故障样本特征进行高效学习,形成故障诊断模型,能够较为准确的识别故障类别。相较于SVM算法,模型精确度提高5.39%,模型鲁棒性提升7.56%。与几种常见机器学习算法的性能进行比较,实验结果表明,使用XGBoost特征提取方法结合简单的分类器可以取得很好的效果。In view of the shortcomings of the commonly used analysis methods of dissolved gas in transformer oil, this paper attempts to use machine learning algorithm to diagnose and identify transformer faults according to a large amount of data generated by oil chromatography online monitoring system. Due to the small size samples in transformer fault diagnosis and the poor generalization ability of machine learning algorithm, in this paper, we propose an extreme gradient lift(XGBoost) algorithm, use the SoftMax function to calculate potential failure warnings, combined with the theorem of gas and fault characteristics in oil, which is screened and constructed, based on feature importance, grid search,cross validation and the parameter optimization model. Taking the dissolved gas content in oil of 334 sets of transformers as an example, XGBoost algorithm can efficiently learn the characteristics of fault samples, form a fault diagnosis model, and identify fault categories more accurately. Compared with SVM algorithm, the accuracy of the model is improved by 5.39%, and the robustness of the model is improved by 7.56%. Compared with the performance of several common machine learning algorithms, the experimental results show that the XGBoost feature extraction method combined with a simple classifier can achieve good results.

关 键 词：变压器 故障诊断 油中溶解气体分析 机器学习 XGBoost 

分 类 号：TM406[电气工程—电器]