摘要
在核电站运行过程中,及时检测故障的发生并进行故障诊断,对核电站的安全稳定运行至关重要。选取核动力一回路系统作为主要研究对象,采用分布式思想,结合深度学习和信息融合的诊断方法,建立了核动力一回路分布式故障诊断系统。首先,通过皮尔逊相关系数以及排序融合的卡方得分和轻量级梯度提升机(light gradient boosting machine,LightGBM)特征重要性进行特征选择。随后,利用深度可分离密集残差网络(deep separable dense residual network,DSDRN)进行故障诊断,通过加深网络深度提取更深层次的特征信息,采用深度可分离卷积减少网络参数量,并通过残差连接降低梯度消失的风险,保证网络训练稳定性。所提模型的各类故障识别准确率均大于98.5%,可为核电站故障诊断工程提供参考。
When the nuclear power plant is in operation,timely detection and diagnosis of faults are crucial for the safe and stable operation of it.Selecting the nuclear power primary loop system as the main research object,adopting a distributed approach combined with deep learning and information fusion techniques for fault diagnosis,a distributed fault diagnosis system for the nuclear power primary loop was developed.Firstly,feature selection was performed based on Pearson correlation coefficient and rank fusion by chi-square score and light gradient boosting machine(LightGBM)feature importance.Subsequently,the deep separable dense residual network(DSDRN)was employed for fault diagnosis.By deepening the network depth to extract deeper feature information,using depthwise separable convolutions to reduce the number of parameters in the network,and reducing the risk of gradient vanishing through residual connections,the stability of network training was ensured.The proposed model achieves accuracy greater than 98.5% in identifying various types of faults,which can provide a reference for fault diagnosis in nuclear power plants.
作者
姬嘉益
马洁
JI Jiayi;MA Jie(College of Mechanical and Electrical Engineering,Beijing Information Science&Technology University,Beijing 100192,China)
出处
《北京信息科技大学学报(自然科学版)》
2025年第4期32-41,共10页
Journal of Beijing Information Science and Technology University(Science and Technology Edition)
基金
国家自然科学基金项目(61973041)。
关键词
核电站
分布式故障诊断
特征选择
残差网络
深度可分离卷积网络
密集块
nuclear power plant
distributed fault diagnosis
feature selection
residual network
depthwise separable convolutional network
dense block
