基于彩色递归图和改进残差网络的核电站主冷却系统故障诊断  

Fault Diagnosis of Nuclear Power Plant Main Cooling System Based on Color Recursion Graph and Improved Residual Network

作　　者：张建红 马洁 袁金晓 王少华 ZHANG Jianhong;MA Jie;YUAN Jinxiao;WANG Shaohua(Mechanical Electrical Engineering School,Beijing Information Science and Technology University,Beijing 100192,China;China Nuclear Power Engineering Co.,Ltd.,Beijing 100142,China)

机构地区：[1]北京信息科技大学机电工程学院,北京100192 [2]中国核电工程有限公司,北京100142

出　　处：《机床与液压》2025年第3期220-227,共8页Machine Tool & Hydraulics

基　　金：国家自然科学基金面上项目(61973041)。

摘　　要：针对压水堆核电站的主冷却系统信号复杂,难以充分提取冷却系统的故障信号,并且神经网络特征在进行特征提取时出现精度较低、提取难度较大、信号部分丢失等情况,提出一种彩色递归图与改进残差密集网络相结合的模型,用于主冷却系统的故障诊断。研究选取5种核电站典型故障。将非线性的核电站一维故障信号转换为二维彩色图片,用于获取故障的详细信息;将二维彩色图像输入到改进的残差密集神经网络进行模型训练。经注意力机制改进的残差密集网络能够提高识别的准确率。最后通过某核电站模拟数据来验证该方法的性能。实验结果以及对比分析表明:该算法可以有效提高主冷却系统的故障识别精度,具有较好的识别效果。For the complex signal of the main cooling system of PWR nuclear power plant,it is difficult to fully extract the fault signal of the cooling system,and the neural network will cause low accuracy,difficulty in extraction and partial loss of signals during feature extraction.To solve this problem,a model combining color recursion graph(CRP)and improved residual dense network was proposed for fault diagnosis of primary cooling system.Five typical faults of nuclear power plant were studied.The nonlinear 1D fault signal of nuclear power plant was converted into 2D color picture,which was used to obtain the detailed fault information.2D color images were input into the improved residual dense neural network for model training.The improved residual dense network can improve the recognition accuracy.Finally,the simulation data of a nuclear power plant were used to verify the performance of the proposed method.The experimental results and comparative analysis show that the algorithm can effectively improve the fault identification accuracy of the main cooling system,and has a good recognition effect.

关 键 词：主冷却系统 故障诊断 彩色递归图 残差密集网络 注意力机制 

分 类 号：TM623[电气工程—电力系统及自动化]