基于多图卷积神经网络的主汽温系统故障诊断  

Fault Diagnosis of Main Steam Temperature System Based on Multi-graph Convolutional Neural Network

作　　者：吴铮 张悦 董泽[1,2] WU Zheng;ZHANG Yue;DONG Ze(School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China;Hebei Technology Innovation Center of Simulation&Optimized Control for Power Generation,Baoding 071003,Hebei Province,China)

机构地区：[1]华北电力大学控制与计算机工程学院,北京102206 [2]河北省发电过程仿真与优化控制技术创新中心,河北保定071003

出　　处：《动力工程学报》2023年第2期237-245,共9页Journal of Chinese Society of Power Engineering

基　　金：河北省省级科技计划资助项目(22567643H);中央高校基本科研业务费专项资金资助项目(2018QN096);河北省自然科学基金资助项目(E2018502111)。

摘　　要：针对主汽温系统具有大迟延、大惯性、非线性和时变性的特点,提出了一种基于多图融合-图卷积神经网络的故障诊断方法。建立邻接图和相关性图,将机组历史运行数据扩展为非欧式空间的图数据,引入特征权重和截断参数来约束节点间的相关性,对图信息进行融合。同时,利用邻接矩阵建立各运行数据间的拓扑信息,并通过深度图卷积结构融合邻近节点信息,建立系统数据与运行状态间的映射关系。结果表明:相较于概率神经网络(PNN)、长短期记忆神经网络(LTSM)和最小二乘支持向量机(LSSVM),所提MG-GCN模型的故障诊断准确率分别提升了11%、7%和16%,误检率、漏检率均较低,能够对多种系统故障类型进行准确识别,具有良好的故障诊断性能。In view of the characteristics of the main steam temperature system with large delay,large inertia,nonlinearity and time-varying,a fault diagnosis method based on multi-graph fusion convolutional neural network was proposed.The adjacency graph and correlation graph were established to expand the historical operation data of the unit into non-Euclidean space graph data,and feature weights and truncation parameters were introduced to constrain the correlation between nodes.At the same time,the adjacency matrix was used to establish the topology information between the operating data,and information of the adjacent nodes was fused through the depth graph convolution structure to establish the mapping relationship between the system data and the operating state.Results show that compared with probabilistic neural network(PNN),long short-term memory(LTSM) and least squares support vector machine(LSSVM),fault diagnosis accuracy of the proposed MG-GCN model has been improved by 11%,7% and 16% respectively,the model has lower false detection rate and missing detection rate,which can accurately identify various system fault types,and has good fault diagnosis performance.

关 键 词：图卷积神经网络 主蒸汽温度系统 故障诊断 火电机组 拓扑结构 

分 类 号：TK39[动力工程及工程热物理—热能工程]