电网N-1下融合CNN与Transformer的综合能源系统静态安全校核  

Static security check of integrated energy system based on fusion of CNN and Transformer under power grid N-1

作　　者：陈厚合[1] 丁唯一 刘光明 李雪[1] 张儒峰 CHEN Houhe;DING Weiyi;LIU Guangming;LI Xue;ZHANG Rufeng(School of Electrical Engineering,Northeast Electric Power University,Jilin 132012,China;Jilin Jiaohe Pumped Storage Co.,Ltd.,State Grid Xinyuan Company,Jilin 132500,China)

机构地区：[1]东北电力大学电气工程学院,吉林吉林132012 [2]国网新源吉林蛟河抽水蓄能有限公司,吉林吉林132500

出　　处：《电力自动化设备》2025年第5期1-9,18,共10页Electric Power Automation Equipment

基　　金：国家自然科学基金资助项目(52077028)。

摘　　要：风光等新能源高比例渗透衍生出大量的源-荷场景,电-气综合能源系统(IEGS)的N-1安全校核面临计算挑战。深度学习技术在处理大量数据时具备显著优势,为解决该问题提供了新的思路。将评价电力系统安全性的Hyper-box和Hyper-ellipse判据推广到天然气系统,并形成IEGS综合安全指标以划分子系统的运行状态;构建卷积神经网络(CNN)-Transformer神经网络以适应量测数据与校核目标的非线性关系,实现快速校核;考虑到系统数据的量纲和数值差异大以及系统状态离散化的特点,分别对数据进行Z-score标准化和独热编码数值化以提升校核精度,并设计改进焦点损失函数以进一步提取不同的场景下天然气系统运行状态的变化规律。以含高比例新能源的综合能源系统(E5G5、E39G20系统)为算例,验证所提方法的高效性和准确性。The high proportion of renewable energy such as wind and solar has derived a large number of source-load scenarios,so the N-1 safety check of integrated electricity-gas system(IEGS)is faced with computational challenges.Deep learning technology has significant advantages in processing large amounts of data and provides a new way to solve this problem.The Hyper-box and Hyper-ellipse criteria for evaluating the safety of power system are extended to the natural gas system,and the comprehensive safety indexes of IEGS are formed to classify the operation states of subsystems.The convolutional neural network(CNN)-Transformer neural network is constructed to adapt to the nonlinear relationship between the measurement data and the check target and realize fast check.Considering the large differences in the dimensions and values of the system data and the characteristics of the discrete state of the system,Z-score standardization and one-hot encoding numeralization of the data are carried out to improve the check accuracy,and a modified focus loss function is designed to further extract the change law of natural gas system’s operation states under different scenarios.The integrated energy systems with a high proportion of renewable energy(E5G5 system and E39G20 system)are taken as the examples to verify the efficiency and accuracy of the proposed method.

关 键 词：电-气综合能源系统 N-1安全校核 深度学习 卷积神经网络 Transformer神经网络 改进焦点损失函数 

分 类 号：TM73[电气工程—电力系统及自动化]