一种可解释人工智能(XAI)在测量设备故障诊断和寿命预测中的应用  

An Application of Explainable Artificial Intelligence(XAI)for Fault Diagnosis and Life Prediction of Measurement Equipment

作　　者：陈长基 梁树华 吴达雷 于秀丽[2] 陈育培 吴孟科 顾婷婷 CHEN Changji;LIANG Shuhua;WU Dalei;YU Xiuli;CHEN Yupei;WU Mengke;GU Tingting(Hainan Power Grid Co.Ltd.,Haikou 570100,China;School of Automation,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]海南电网有限责任公司,海口570100 [2]北京邮电大学自动化学院,北京100876

出　　处：《西南大学学报（自然科学版）》2024年第1期167-177,共11页Journal of Southwest University(Natural Science Edition)

基　　金：国家重点科技项目(2020YFB0906003);中国南方电网有限责任公司科技项目(070000KK52200021).

摘　　要：基于人工智能算法的变压器故障诊断和寿命预测模型在提高准确率方面已经达到了很好的效果,但是仍存在泛化性能较低,对数据质量要求过高,判断结果无法解释等问题.该文基于DBSO-CatBoost模型,提出一种可用于故障判断解释的变压器故障诊断方法.该方法基于数据特征提取,采用差分变异头脑风暴优化(DBSD)算法对CatBoost模型进行优化和故障诊断.①对于数据预处理,引入比率法在原始数据中添加特征;采用基于可解释人工智能(XAI)的Shapley加法解释(SHAP)技术进行特征提取,并采用核主成分分析算法对数据进行降维.Shapley加法解释技术可根据特征贡献解码每个预测来帮助全局解释并评估预测结果.②将预处理后的数据输入到CatBoost模型中进行训练,并采用差分变异头脑风暴优化算法对CatBoost模型的参数进行优化,从而得到最优模型.③利用得到的优化模型诊断变压器故障并输出故障类型与预测结果.实验使用来自中国国家电网公司西北部某电网的真实数据评估该模型.结果表明:该文模型在不同故障诊断中的准确性最佳,平均准确率高达99.29%,证明该文方法可以有效提高电力变压器故障诊断的准确性和效率.The current transformer fault diagnosis and life prediction models based on artificial intelligence algorithms have achieved good results in improving the accuracy,but there are still problems such as low generalization performance,excessive requirements for data quality,and uninterpretable judgment results.To this end,this paper proposes a transformer fault diagnosis method based on DBSO-CatBoost model that can be used for fault judgment interpretation.The method is based on data feature extraction,and using the differential variance brainstorming optimization algorithm to optimize the CatBoost model and diagnose the fault.①For data preprocessing,the ratio method was introduced to add features to the original data.The Shapley additive interpretation(SHAP)technique based on explainable artificial intelligence(XAI)was used for feature extraction,and the kernel principal component analysis algorithm was used to reduce the dimensionality of the data.The Shapley additive interpretation technique can decode each prediction based on the feature contribution to help global interpretation and evaluate the prediction results.②The preprocessed data was fed into the CatBoost model for training and the parameters of the CatBoost model were optimized using the difference variance brainstorming optimization algorithm to obtain the optimal model.③The obtained optimal model was used to diagnose transformer faults and output the fault type and prediction results.The experiments evaluated the model using real data from a power grid in the northwest of the State Grid Corporation of China.The results show that the model of this paper has the best accuracy in different fault diagnosis,with an average accuracy of 99.29%.It is proved that the proposed method can effectively improve the accuracy and efficiency of power transformer fault diagnosis.

关 键 词：可解释人工智能 故障诊断 寿命预测 机器学习 电力变压器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]