电力变压器油中溶解气体异常数据识别与含量预测研究  

Research on Abnormal Data Identification and Content Prediction of Dissolved Gas in Power Transformer Oil

作　　者：杜江[1,2] 范志远 范仲华 王庆凯 李佩贤 DU Jiang;FAN Zhiyuan;FAN Zhonghua;WANG Qingkai;LI Peixian(Key Laboratory of Electromagnetic Field and Electrical Apparatus Reliability of Hebei Province(Hebei University of Technology),Beichen District,Tianjin 300401,China;State Key Laboratory of Reliability and Intelligence of Electrical Equipment(Hebei University of Technology),Beichen District,Tianjin 300401,China;State Grid Shandong Xiajin Power Supply Company,Dezhou 253200,Shandong Province,China)

机构地区：[1]河北省电磁场与电器可靠性重点实验室(河北工业大学),天津市北辰区300401 [2]省部共建电工装备可靠性与智能化国家重点实验室(河北工业大学),天津市北辰区300401 [3]国网山东夏津县供电公司,山东省德州市253200

出　　处：《电网技术》2025年第2期844-853,I0116,共11页Power System Technology

基　　金：国家自然科学基金项目(52007047);天津市自然科学基金重点项目(19JCZDJC32100)。

摘　　要：采用神经网络模型对油中溶解气体含量进行预测是目前评估电力变压器运行状态的重要方法,数据质量是影响神经网络模型预测精度的关键因素,然而,由于变压器复杂的运行环境,使得采集到的气体数据中不可避免地存在多种类型的异常数据,进而造成数据质量下降,严重影响模型的预测精度。此外,神经网络模型的参数是否匹配也是影响其预测性能的重要因素,然而,传统依据人工经验选择参数存在主观性、低效性和不可扩展性等缺点,也在一定程度上影响了模型的预测性能。为解决上述问题,该文通过对最近邻集成隔离法(isolation using nearest neighbor ensemble,iNNE)进行修正,提出了修正最近邻集成隔离法(modified isolation using nearest neighbor ensemble,MiNNE),利用MiNNE综合考虑局部度量与全局度量的特性实现气体异常数据的准确识别,有效提升数据质量。同时,对鹈鹕优化算法进行改进,提出了改进鹈鹕优化算法(improved pelican optimization algorithm,IPOA),并利用IPOA对影响神经网络模型预测精度的关键参数进行优化,有效克服了传统依据经验选参而导致模型预测精度低与传统POA易陷入局部最优的问题,提高了模型的预测性能。采用电力变压器实际运行数据对所提模型进行验证,结果表明,相较于其他模型,所提模型在7种特征气体预测中均取得了最佳的预测效果,充分证明了所提模型的优越性。Using a neural network model to predict the content of dissolved gas in oil is an important method to evaluate the operation state of a power transformer,and the data quality is the key factor affecting the prediction accuracy of the neural network model.However,due to the complex operating environment of the transformer,there are inevitably many types of abnormal data in the collected gas data,which leads to the decline of data quality.It seriously affects the prediction accuracy of the neural network model.In addition,whether the parameters of the neural network model match or not is also an important factor affecting its prediction performance.However,the traditional selection of parameters based on artificial experience has shortcomings,such as subjectivity,low efficiency,and expansibility,affecting the model's prediction performance to a certain extent.To solve the above problems by modifying the isolation using nearest neighbor ensemble(iNNE),this paper proposes a modified isolation using nearest neighbor ensemble(MiNNE),which uses the characteristics of MiNNE to comprehensively consider local and global metrics to accurately identify gas anomaly data and effectively improve the data quality.At the same time,the pelican optimization algorithm is improved,and an improved pelican optimization algorithm(IPOA)is proposed,and IPOA is used to optimize the key parameters that affect the prediction accuracy of the neural network model,which effectively overcomes the problems of low prediction accuracy of the model caused by traditional empirical parameter selection and the traditional POA easy to fall into local optimization,and improves the prediction performance of the model.The actual operation data of the power transformer verify the model proposed in this paper.The results show that,compared with other models,this model has achieved the best prediction results in the prediction of seven kinds of characteristic gases,which fully proves the superiority of this model.

关 键 词：电力变压器 油中溶解气体 异常数据识别 气体含量预测 

分 类 号：TM721[电气工程—电力系统及自动化]