基于门控循环网络的故障诊断方法研究  

Research on Fault Diagnosis Method Based on Gate Recurrent Unit Network

作　　者：韩东江 戴新发[1] HAN Dongjiang;DAI Xinfa(Wuhan Digital Engineering Institute,Wuhan 430205)

机构地区：[1]武汉数字工程研究所,武汉430205

出　　处：《舰船电子工程》2024年第10期79-84,共6页Ship Electronic Engineering

摘　　要：核动力装置具有内反馈和强耦合的特征,使得内部工况环境复杂,一旦出现异常则更有可能发生高风险事故。为了提高核动力装置运行的安全性,论文提出了一种基于门控循环网络的故障诊断方法,该方法利用核动力装置运行过程中产生的数据进行瞬态特征与时变特征的提取,通过梯度提升树算法和门控循环网络算法进行联合诊断。实验结果表明,与其他模型相比,该模型能够更加准确地判断每一个故障,故障诊断的准确率达到99.9%以上。Nuclear power plant has the characteristics of internal feedback and strong coupling,which makes the internal working environment complex,and once the abnormal situation is more likely to occur high-risk accidents.In order to improve the safety of nuclear power plant operation,a fault diagnosis method based on gated cycle network is proposed in this paper,in this method,the transient and time-varying features are extracted from the data generated during the operation of nuclear power plant.The experimental results show that the model can judge each fault more accurately than other models,and the accuracy of fault diagnosis is over 99.9%.

关 键 词：核反应堆装置 故障诊断 梯度提升树 门控循环网络(GRU) 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]