一种基于夏普利值及油中溶解气体分析的可解释变压器故障诊断方法  

Interpretable Transformer Fault Diagnosis Based on SHAP Value and Dissolved Gas Analysis of Transformer Oil

作　　者：廖才波 杨金鑫 邱志斌 胡雄 曾清霖 黄智勇[1] LIAO Caibo;YANG Jinxin;QIU Zhibin;HU Xiong;ZENG Qinglin;HUANG Zhiyong(Department of Energy and Electrical Engineering,Nanchang University,Nanchang 330031,Jiangxi Province,China;Ganzhou Power Supply Company,State Grid Jiangxi Electric Power Co.,Ltd.,Ganzhou 341000,Jiangxi Province,China)

机构地区：[1]南昌大学能源与电气工程系,江西省南昌市330031 [2]国网江西省电力有限公司赣州供电分公司,江西省赣州市341000

出　　处：《电网技术》2024年第4期1752-1761,I0099,共11页Power System Technology

基　　金：国家自然科学基金项目(62163025);江西省自然科学基金项目(20212ACB212007)。

摘　　要：相比于三比值等传统方法,基于机器学习算法的变压器故障诊断方法在诊断效率及准确性等方面具有一定的优势,但“黑箱模型”的本质属性决定了其决策过程及诊断结果的不可解释性。针对该问题,该文提出了一种基于油中溶解气体分析的可解释变压器故障诊断方法,采用树形夏普利加法解释(tree Shapely additive explanations,TreeSHAP)方法实现了基于树结构概率密度估计优化极端梯度提升(tree-structured parzen estimator-extreme gradient boosting,TPE-XGBoost)的变压器故障诊断模型的可解释性分析。首先,构建了涵盖油中溶解气体含量、比值及编码等多结构数据的24维故障特征集合,并筛选得到10个有效特征量。其次,提出了基于TPE-XGBoost算法的变压器故障诊断方法,采用树结构概率密度估计完成XGBoost模型的多参数同步优化,实现对故障类型的准确判断。最后,引入TreeSHAP理论开展变压器故障诊断模型的可解释性分析,实现了故障诊断决策过程及其影响因素的可视化,并获取了不同故障类型的关键特征量。研究表明,该文所述变压器故障诊断方法的平均准确率为90.23%,同时可反映特征量对模型决策的影响过程及程度。该方法具有较好的准确性、鲁棒性及可解释性,可为变压器运维检修提供针对性的指导建议。Compared with the traditional methods such as the Three Ratios,the methods for transformer fault diagnosis based on the machine learning algorithms have the diagnostic efficiency and accuracy advantages.However,the inherent attribute of the"black-box model"determines the inexplicability of the decision-making process and the diagnostic results.To solve this problem,an interpretable transformer fault diagnosis based on the dissolved gas analysis of the transformer oil is proposed in this paper.The tree shapely additive explanations(TreeSHAP)method is adopted to realize the interpretability analysis of the fault diagnosis model based on the tree-structured parzen estimator-extreme gradient boosting(TPE-XGBoost).First,a 24-dimensional fault feature set covering the multi-structural data such as the content,the ratio,and the code of the dissolved gas in the oil is constructed,and 10 available features are picked out and obtained.Then,the transformer fault diagnosis model based on the TPE-XGBoost is proposed.The tree-structured parzen estimator is used to complete the multi-parameter synchronous optimization of the XGBoost model,realizing the accurate judgment of the fault types.Finally,the TreeSHAP theory is introduced to analyze the interpretability of the transformer fault diagnosis model,achieving the visualization of the fault diagnostic decision-making process and its influencing factors and obtaining the key features of different fault types.The research shows that the average accuracy of the transformer fault diagnosis proposed in this paper reaches 90.23%,and the influence process and degree of the features on the model decision-making can be reflected.This method has good accuracy,robustness,and interpretability,which will provide targeted guidance and suggestions for the transformer operation and maintenance.

关 键 词：变压器 故障诊断 树结构概率密度估计 极端梯度提升 可解释性 夏普利值 

分 类 号：TM721[电气工程—电力系统及自动化]