基于机器学习的±800 kV干式平波电抗器震后状态评估方法  

Post⁃seismic State Assessment Method for±800 kV Dry⁃type Smoothing Reactor Based on Machine Learning

作　　者：叶芳怡 刘匀[1] 朱旺 谢强[1] YE Fangyi;LIU Yun;ZHU Wang;XIE Qiang(College of Civil Engineering,Tongji University,Shanghai 200092,China)

机构地区：[1]同济大学土木工程学院,上海200092

出　　处：《高压电器》2025年第4期1-11,共11页High Voltage Apparatus

基　　金：国家自然科学基金资助项目(51878508)。

摘　　要：为实现电气设备的震后状态快速评估,以保障电力系统的应急决策与灾后恢复顺利进行,文中提出一种基于机器学习的评估方法。以±800 kV干式平波电抗器为研究对象,建立Abaqus有限元模型进行地震响应分析确定其抗震薄弱环节。向有限元模型中输入大量地震波计算获得用于建立评估模型所需的机器学习数据集,采取相关性分析法剔除冗余特征,选取不同机器学习算法比较其评估性能,并通过SHAP(shapley additive explanations)解释评估模型,以此避免机器学习模型的“黑箱”特性。结果表明:平波电抗器的抗震薄弱环节为支撑绝缘子根部应力响应;基于XGBoost算法架构的评估模型具有最优性能;SHAP法可从全局和局部层面揭示地震动参数对设备震后状态的影响。基于机器学习算法建立的评估模型能够快速准确地评估设备震后状态,可为变电站或换流站整站的智能化防灾系统建立提供技术支撑。For achieving rapid post-seismic state assessment of electrical equipment and ensuring effective emergen-cy decision-making and post-disaster recovery of power systems,a kind of machine learning-based assessment meth-od is proposed in this paper.The±800 kV dry-type smoothing reactor is taken as the research object,and Abaqus fi-nite element model is set up for seismic response analysis so to determine its weak seismic points.A large number of seismic waves are input into the finite element model to obtain the machine learning data set required for setting up the assessment model.The redundant features are removed by using correlation analysis and its assessment perfor-mance is compared by selecting different machine learning algorithms.The shapley additive explanations(SHAP)is used to explain the assessment model so to avoid the black-box characteristics of machine learning model.The re-sults show that the weak seismic point of the smoothing reactor is in the stress response at the root of the supporting insulator.The assessment model based on the XGBoost algorithm possesses the optimal performance.The SHAP method can effectively reveal the influence of seismic parameters on the post-seismic state at both global and local levels.The assessment model setting up based on the machine learning algorithm can quickly and accurately assess post-seismic state of the equipment and provide technical support for the establishment of intelligent disaster preven-tion systems of either substation or converter station.

关 键 词：±800 kV干式平波电抗器 震后评估 机器学习 SHAP法 

分 类 号：TM47[电气工程—电器]