基于多模型选择性融合的变压器在线故障诊断  

On-line Fault Diagnosis for Transformer Based on Selective Fusion of Multiple Models

作　　者：张育杰 李典阳 冯健[1] 王善渊 ZHANG Yujie;LI Dianyang;FENG Jian;WANG Shanyuan(College of Information Science and Engineering,Northeastern University,Shenyang 110819,China;State Grid Liaoning Electric Power Supply Co.,Ltd.,Shenyang 110006,China)

机构地区：[1]东北大学信息科学与工程学院,辽宁省沈阳市110819 [2]国网辽宁省电力有限公司,辽宁省沈阳市110006

出　　处：《电力系统自动化》2021年第13期95-101,共7页Automation of Electric Power Systems

基　　金：国家自然科学基金资助项目(61673093)。

摘　　要：在线智能化故障诊断是变压器故障诊断的发展趋势,为发挥不同征兆子集及诊断算法的性能优势,实现多元诊断模型的互补决策,文中通过对智能诊断算法的性能进行多维对比分析,选取综合性能优的诊断算法用于征兆子集优选,使用优选子集与诊断算法训练多个诊断模型,通过模型间差异性分析有选择地对模型进行融合并用于最终决策。实例验证结果表明,融合模型泛化性能及均衡诊断能力均有提升。Online intelligent fault diagnosis is a development trend of transformer fault diagnosis. In order to make use of the performance advantages of different symptom subsets and diagnosis algorithms, and realize the complementary decision-making of multiple diagnosis models, this paper conducts multi-dimensional comparative analysis on the performance of intelligent diagnosis algorithms, and selects the diagnosis algorithms with excellent comprehensive performance for the selection of the symptom subset. The selected subsets and diagnosis algorithms are used to train multiple diagnosis models, and the models are selectively fused and used for final decision-making through the analysis on the differences between models. The example verification results show that the generalization performance and balance diagnosis ability of the fusion model are improved.

关 键 词：电力变压器 故障诊断 征兆排序 子集优选 模型融合 

分 类 号：TM41[电气工程—电器]