核电站冷却水泵的小样本迁移学习故障诊断模型

打开文本图片集

关键词：冷却水泵；小样本；注意力门控循环神经网络；迁移学习；故障诊断精度DOI：10. 15938/j. jhust. 2025.05.006中图分类号：TH212;TH213.3 文献标志码：A 文章编号：1007-2683（2025）05-0058-12

Abstract：Forthe smallsample imbalance problem inthediagnosis ofcoling water pumpfaults innuclearpower plants，asmall samplefault diagnosis modelforcoling waterpumpsinnuclear powerplants（GRU-AM-TL）thatintegrates transferlearningand attentiongatedrecuentneuralnetworksisproposedFirstly，anaentionlayerisitroducedintotheGatedRecurrntUnit（GRU） neuralnetworktoostuctanatentiongatedecuenteuralnetworktatadaptivelyssigsdirentweightstteextracedfeatures. Then，transferlearning（TL）isusedtolearnknowledgefromrelevantsourcedatasets.Intheexperiment，theGRU-AM-TLmodelwas compared with GRU-based methodsandotherclasicalmethodsusingbearing datasets，gearboxfaultdatasets，andnuclearpowerplant simulated fault datasets.The results showed that the diagnostic accuracy of this method was improved by 1.2% ，9.4%， 12.3% ， 4.8% ，and 2.9% compared to other methods using 2% datasets A～E，reaching 83. 1% ，85.8%，82.8%， 86.4% ，and 83.3% ， respectively，which can effectively improve the fault diagnosis accuracy in small sample scenarios.

Keywords：cooling waterpumps；smallsample；atention gatedrecurrent neural network；transferlearing;fault diagnosis accu racy

0 引言

反应堆冷却水泵等旋转机械是核电站的一种重要设备，具有流体输送、能量传递和转换等重要功能。（剩余16214字）