基于改进双深度Q网络在变压器不平衡样本故障诊断的研究

打开文本图片集

中图分类号：TP39；TP183；TM41 文献标识码：A文章编号：2096-4706（2025）21-0034-05

Research on Transformer Unbalanced Sample Fault Diagnosis Based on Imbalanced Classification Double Deep Q Network

（State Grid Xi'an Electric Power Supply Company，Xi'an 71oo32， China）

Abstract：To address the problem of model bias caused by unbalanced distribution of sample categories in transformer fault diagnosis，this paper proposesan Imbalanced Clasification Double DeepQNetwork （ICDDQN）.Bydesigningadynamic categoryweightreward functionandcombiningKLdivergence toconstructasampledistributioncompensationmechanism，the contributionofdiferentcategoriesofsamplesisdynamicallyadjustedduringthemodeltrainingphase.Experimentsarecaried outbasedonaself-builtdatasetcontaining320Osetsofoilchromatographicdata.ThecomparisonresultsshowthattheFl-score convergence accuracy of ICDDQN reaches 99.25% ，which is superior to the DDQN baseline，and provides technical support for power equipment diagnosis under class imbalance conditions.

Keywords： transformer; fault diagnosis; Deep Reinforcement Learning; Neural Network

0 引言

变压器作为电力系统能量转换的核心枢纽设备，其运行可靠性直接关系电网安全[。（剩余5207字）