哈尔滨商业大学学报（自然科学版）2025，Vol.41Issue(6)：691-699,9.
基于CNN-LSTM-KAN的核电厂一回路故障诊断方法研究
Study on fault diagnosis method of nuclear power plant primary loop systems based on CNN-LSTM-KAN
摘要
Abstract
Data-driven fault diagnosis techniques are significant for the stable operation of nuclear power plants(NPPs).In this paper,the primary loop system of a nuclear power plant was selected as the research object.To enhance its fault diagnosis capability,an integrated fault diagnosis model combining a Convolutional Neural Network(CNN),Long Short-Term Memory(LSTM),and Kolmogorov-Arnold Network(KAN)was proposed.In this method,CNN was employed to extract implicit static features from monitoring data,while LSTM was utilized to capture dynamic characteristics along the temporal dimension.Subsequently,the fully connected layers in the model were replaced with KAN to further improve nonlinear fitting capability and generalization performance.Experiments were conducted using a typical fault dataset generated by a nuclear power plant simulator.The proposed CNN-LSTM-KAN model was compared with LSTM-based and CNN-LSTM-based models across datasets with different fault severity levels.Results demonstrated that the CNN-LSTM-KAN model achieved a high fault identification accuracy of 99.39%,exhibited excellent performance under various fault conditions,and possessed strong generalization capability.关键词
压水反应堆/一回路系统/故障诊断/CNN/LSTM/Kolmogorov-Arnold网络Key words
power water reactor/primary loop system/fault diagnosis/CNN/LSTM/Kolmogorov-Arnold Network分类
信息技术与安全科学引用本文复制引用
顾阳,刘永阔,张万洲,单龙飞,刘极..基于CNN-LSTM-KAN的核电厂一回路故障诊断方法研究[J].哈尔滨商业大学学报（自然科学版）,2025,41(6):691-699,9.基金项目
国家自然科学基金联合基金重点项目(U21B2083) （U21B2083）
