基于GA-BP神经网络的SF_(6)/N_(2)混合气体GIS设备故障诊断  

GA-BP Neural Network-based Fault Diagnosis GIS Equipment with SF_(6)/N_(2) Gas Mixture

作　　者：梁璐[1] 蒋延磊[1] 曹心怡 苏鑫 郑俊洋[1] 丁五行 LIANG Lu;JIANG Yanlei;CAO Xinyi;SU Xin;ZHENG Junyang;DING Wuxing(State Grid Henan Electric Power Company,Pingdingshan 467000,China;T&P Union(Beijing)Co.,Ltd.,Beijing 100096,China)

机构地区：[1]河南省电力公司平顶山供电公司,河南平顶山467000 [2]泰普联合科技开发(北京)有限公司,北京100096

出　　处：《电工技术》2024年第13期27-31,共5页Electric Engineering

基　　金：国网河南省电力公司科技项目“基于分解物与局放检测的SF6/N2混合气体故障诊断技术研究”(编号521760230002)。

摘　　要：为实现SF_(6)/N_(2)混合气体GIS设备的潜伏性故障诊断,搭建实验平台探究SO_(2)浓度与放电量之间的关系,以SO_(2)浓度增量为输入参数之一,建立BP、GA-BP神经网络模型来预测四种缺陷类型下的放电量,分析预测结果后选取最优诊断算法。结果表明:每种缺陷类型下的放电量与SO_(2)浓度之间呈现正相关性,GA-BP神经网络模型的预测准确率和拟合度分别为0.86和0.9386,平均相对误差为1.23%,在评估结果中占明显优势,可为SF_(6)/N_(2)混合气体GIS设备的潜伏性故障建立诊断算法提供基础性数据。In order to diagnose latent faults in SF_(6)/N_(2) gas mixture GIS equipment,an experimental platform was constructed to explore the relationship between the content of SO_(2) gas detected by ultraviolet fluorescence method and the discharge quantity.BP and GA-BP neural networks were established to predict the discharge amount under four typical defect types,and the increase in SO_(2) concentration was added as one of the input parameters.The predictive results were compared and analyzed to select the optimal diagnostic algorithm.The results show that there is a positive correlation between the discharge quantity and the concentration of SO_(2) for each type of discharge defect.The prediction accuracy and fitting degree of the GA-BP neural network model are 0.86 and 0.9386 respectively,with an average relative error of 1.23%.The proposed method has a significant advantage with respect to evaluation results and can provide fundamental data for establishing a diagnostic algorithm for latent faults in SF_(6)/N_(2) gas mixture GIS equipment.

关 键 词：SF_(6)/N_(2)混合气体 SO_(2)浓度 GA-BP神经网络模型 潜伏性故障 

分 类 号：TM59[电气工程—电器]