电力系统及其自动化学报2025，Vol.37Issue(10)：120-128,9.DOI:10.19635/j.cnki.csu-epsa.001590
配电网故障识别Transformer-联邦迁移学习算法设计
Design of Transformer-based Federated Transfer Learning Algorithm for Distribution Network Fault Recognition
摘要
Abstract
The recognition of fault types in distribution networks is crucial for enhancing the reliability of power supply.At present,artificial intelligence algorithms have a broad application prospect in the field of power system fault diagno-sis.In this paper,a Transformer-based federated transfer learning algorithm is proposed for fault recognition in distribu-tion networks.Specifically,by enhancing the Transformer encoder module,the global information within the input of electrical signals is effectively captured,enabling more accurate and efficient fault recognition.Then,federated learn-ing is used to pass the client model parameters,which solves the problems of data errors,delays or mis-coding during the communication process.In addition,through deep transfer learning,the model can utilize the training results from one line to effectively classify the fault types on other lines within the same distribution network.Simulation results show that the proposed algorithm can outperform the conventional algorithm in the case of high-resistance faults,and it also has a good model generalization capability.关键词
配电网/故障类型辨识/联邦学习/深度迁移学习/Transformer网络Key words
distribution network/fault type recognition/federated learning/deep transfer learning/Transformer net-work分类
动力与电气工程引用本文复制引用
叶远波,王吉文,邵庆祝,汪艳霞..配电网故障识别Transformer-联邦迁移学习算法设计[J].电力系统及其自动化学报,2025,37(10):120-128,9.基金项目
国家电网公司科技项目(521200223001N). （521200223001N）
