基于TimeGAN的轨道交通LTE-M故障预测研究

打开文本图片集

中图分类号：TP391 文献标识码：A 文章编号：2096-4706（2025）08-0010-06

Abstract：The Long TermEvolution of Metro （LTE-M） network fault prediction datasetof rail transit has the problems ofunbalancedsamplesandsmallamountofsampledatawhichimpacttheacuracyoffault prediction.Inordertosolvethe above problems，this paper proposes aresearch methodofLTE-Mfault predictionofrail transitbased onconditionalTime-series Generative Adversarial Networks （TimeGAN）.Bydefiningdynamic autoencoderandstatic autoencoderinTimeGANmodel，this method furtherexploresthedynamicandstaticcharacteristicsofLT-Mfaultdataofrailtransit，andintroducesGELUactivation functionnthepotentialspaceofgeneratoranddiscriminatortoaceleratemodelconvergenceandgeneratesyntheticdatacloser toreal data，thusefectivelyalleviating the problemofunbalancedfaultdatasetandsmalldatavoume.Theexperimentalresults showthatwhenthedatasynthesizedbytheTimeGANmodelisusedforfaultpredictiontraining，itcanproducebeterediction results than the original data.

Keywords：rail transitLTE-M;fault prediction;time-series;TimeGAN

0 引言

随着新一代移动通信的飞速发展，轨道交通通信基础设施规模也迅速扩展，LTE-M网络作为轨道交通网络关键组成部分，其复杂性也随之增加。（剩余7847字）