基于深度迁移学习的电力系统暂态状态估计  

Transient State Estimation for Power System Based on Deep Transfer Learning

作　　者：焦昊 赵佳伟 韦磊 朱卫平 马洲俊 臧海祥[2] JIAO Hao;ZHAO Jiawei;WEI Lei;ZHU Weiping;MA Zhoujun;ZANG Haixiang(Electric Power Research Institute,State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 211103,China;School of Electrical and Power Engineering,Hohai University,Nanjing 211100,China;State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 210024,China)

机构地区：[1]国网江苏省电力有限公司电力科学研究院,南京市211103 [2]河海大学电气与动力工程学院,南京市211100 [3]国网江苏省电力有限公司,南京市210024

出　　处：《电力建设》2025年第1期97-106,共10页Electric Power Construction

基　　金：国网江苏省电力有限公司科技项目(J2023121)。

摘　　要：电力系统故障样本数据稀缺导致电力系统暂态情况下的状态难以实时、准确跟踪,针对此问题提出一种基于深度迁移学习的电力系统暂态状态估计方法。首先,依托数字孪生技术得到实际电力系统运行的孪生数据,为暂态状态估计提供充足的样本数据来源。然后,将孪生数据划分为源域数据集和目标域数据集,并基于电力系统稳态数据构建源域状态估计基础模型。最后,通过深度迁移学习技术,利用目标域中小样本暂态数据来微调基础模型,即可获得适用于暂态的状态估计模型,提升状态估计器的普适性。仿真表明,当电力系统发生故障时,所提方法与不使用迁移学习的深度神经网络相比具有更高的估计精度和计算效率。A method for transient state estimation in power systems based on deep transfer learning is proposed to accurately track transient state in real-time,which is typically challenging owing to the limited availability of fault sample data.Initially,the twin data representing the actual power system operation are generated by utilizing digital twin technology,thereby providing substantial sample data sources for transient state estimation.Subsequently,the twin datasets are partitioned into source domain and target domain datasets,and a base model is developed for state estimation in the source domain based on steady-state power system data.Finally,by applying deep transfer learning,the base model is fine-tuned using small-sample transient data in the target domain,resulting in a state-estimation model specifically adapted for transient conditions and enhancing the universality of the estimator.Simulations demonstrate that the proposed method exhibits a higher estimation accuracy and computational efficiency than that of deep neural networks without transfer learning,particularly during power system failures.

关 键 词：电力系统故障 数字孪生 暂态状态估计 深度迁移学习 小样本 

分 类 号：TM73[电气工程—电力系统及自动化]