内蒙古电力技术2025，Vol.43Issue(3)：32-38,7.DOI:10.19929/j.cnki.nmgdljs.2025.0032
基于TICC算法的自监督学习核电设备运行工况划分
Operating Condition Classification of Self-Supervised Learning Nuclear Power Equipment Based on TICC Algorithm
摘要
Abstract
In order to realize the accurate,rapid and automatic division of the operating conditions of nuclear power equipment,conduct health assessment of nuclear power equipment,and detect equipment abnormalities in time,a self-supervised learning algorithm for the division of operating conditions of nuclear power equipment based on Toeplitz inverse covariance-based clustering(TICC)algorithm is proposed.Firstly,the historical operation data of nuclear power equipment are normalized,and the optimal clustering number is determined by elbow method.Then,the TICC algorithm is used to classify the working conditions for the historical operation data of nuclear power equipment,and the working condition labels are assigned to the data fragments of each working condition through the classification results.Finally,the convolutional neural network is trained by using the labeled condition data to obtain the condition division model,and the proposed algorithm is verified by using the real operation data of nuclear power equipment.The results show that the accuracy of the proposed algorithm can reach 96.6%,and the working condition division takes only 3.4 s.Compared with K-means clustering algorithm and TICC algorithm,the algorithm proposed in this paper has a great improvement in accuracy and division speed,which can effectively help the on-site operation and maintenance personnel of nuclear power station to complete the division of the operating conditions of nuclear power equipment.关键词
核电设备/运行工况划分/托普利兹逆协方差聚类算法/K-means聚类算法/卷积神经网络Key words
nuclear power equipment/operating conditions division/Toeplitz inverse covariance-based clustering(TICC)algorithm/K-means clustering algorithm/convolutional neural network分类
信息技术与安全科学引用本文复制引用
张大志,郑胜,崔文浩..基于TICC算法的自监督学习核电设备运行工况划分[J].内蒙古电力技术,2025,43(3):32-38,7.基金项目
中核核工业仿真技术重点实验室对外开放基金项目"基于半监督/自监督的深度学习异常检测方法研究"(B220631) （B220631）
