机械与电子2025，Vol.43Issue(11)：54-60,7.
基于深度学习的牵引网短路电流辨识方法研究
Research on Deep Learning-based Short-circuit Current Identification Method for Traction Network
摘要
Abstract
To resolve misjudgment in traction network protection due to challenges in distinguishing remote faults from maximum load currents,this paper proposes a CNN-BiLSTM-MHA model.The framework featuring space,time sequence and frequency spectrum integrates multimodal fusion and dy-namic weighting:CNN concurrently extracts load current harmonics and fault transients;BiLSTM captures bidirectional fault propagation delays;MHA decouples steady-state loads from transient faults while sup-pressing harmonics and noise.Experiments show 98.73%TR/FR fault identification accuracy and 93.11%robustness in 30～100 dB noise,demonstrating a high-precision and real-time protection solution for in-telligent protection of traction networks.关键词
牵引网/短路电流辨识/卷积神经网络/双向长短期记忆网络/多头注意力机制Key words
traction network/fault current identification/CNN/BiLSTM/MHA分类
信息技术与安全科学引用本文复制引用
李喆,李向学,马贵荣,白亚栋,常宇健..基于深度学习的牵引网短路电流辨识方法研究[J].机械与电子,2025,43(11):54-60,7.基金项目
国能朔黄铁路发展有限责任公司科研课题(SHTL-24-32) （SHTL-24-32）
