Article Details

Abstract

The Long Term Evolution of Metro (LTE-M) network fault prediction dataset of rail transit has the problems of unbalanced samples and small amount of sample data, which impact the accuracy of fault prediction. In order to solve the above problems, this paper proposes a research method of LTE-M fault prediction of rail transit based on conditional Time-series Generative Adversarial Networks (TimeGAN). By defining dynamic autoencoder and static autoencoder in TimeGAN model, this method further explores the dynamic and static characteristics of LTE-M fault data of rail transit, and introduces GELU activation function in the potential space of generator and discriminator to accelerate model convergence and generate synthetic data closer to real data, thus effectively alleviating the problem of unbalanced fault dataset and small data volume. The experimental results show that when the data synthesized by the TimeGAN model is used for fault prediction training, it can produce better prediction results than the original data.

Keywords

rail transit LTE-M; fault prediction; time-series; TimeGAN

References

1. [1] Chen Dongyang, Mao Li Research on Stock Price Prediction by Integrating Incremental Learning and Transformer Model [J]. Computer Science and Exploration, 2024, 18 (7): 1889-1899
2. [2] Li Na, Yang Zhao, Wang Yeping, etc CNN BiGRU quadcopter trajectory prediction method integrating temporal attention [J/OL]. Journal of Wuhan University of Technology: Transportation Science and Engineering Edition, 2025:1-9 [March 13, 2025] http://kns.cnki.net/kcms/detail/42.1824.U.20240409.1221.046.html.
3. [3] Xie Bocai, Palace Lord A review of machine learning based research on turnout fault diagnosis and prediction [J]. Railway Communication Signal Engineering Technology, 2021, 18 (8): 93-99
4. [4] Shi Qingyi, Wang Wei, An Siguang, etc Electrical data generation method based on temporal generative adversarial network and attention mechanism [J]. Modern Electronic Technology, 2024, 47 (11): 161-167
5. [5] Li Lan, Zhang Jie, Liu Jie, etc Social and Scene Aware Pedestrian Trajectory Prediction Based on GAN [J]. Computer Applications and Software, 2024, 41 (6): 72-78
6. [6] ZHANG G Q，GUO J F. A Novel Ensemble Method for Residential Electricity Demand Forecasting Based on A Novel Sample Simulation Strategy [J].Energy，2020：207：118265.
7. [7] CHEN Z S，HOU K R，ZHU M Y，et al. A Virtual Sample Generation Approach Based on a Modified Conditional GAN and Centroidal Voronoi Tessellation Sampling to Cope With Small Sample Size Problems: Application to soft Sensing for Chemical Process [J].Applied Soft Computing，2021，101：107070.
8. [8] Ding Linlin, Hu Yongliang, Li Yuda, etc Transformer coal mine microseismic localization method based on conditional adversarial enhancement [J]. Computer and Digital Engineering, 2024, 52 (1): 1-8+17
9. [9] YOON J，JARRETT D，SCHAAR M V D. Time-Series Generative Adversarial Networks [EB/OL].[2024-09-28].https://papers.nips.cc/paper/8789-time-series-generative-adversarial-networks.pdf.
10. [10] ZHANG Y F，ZHOU Z H，LIU J W. Data Augmentation for Improving Heating Load Prediction of Heating Substation Based on TimeGAN [J].Energy，2022，260：1-12.
11. [11] Wang Yuhong, He Qiduo, Zheng Zongsheng, etc Transient Voltage Stability Prediction Based on Conditional Generative Adversarial Networks and Transfer Learning [J]. Power Automation Equipment, 2025, 45 (2): 159-166
12. [12] Zhou Linru, Peng Pengfei Research on Task Analysis Method Based on Attention Generative Adversarial Networks [J]. Computer Science, 2024, 51 (3): 63-71
13. [13] HARFORD S，KARIM F，DARABI H. Generating Adversarial Samples on Multivariate Time Series using Variational Autoencoders [J].IEEE/CAA Journal of Automatica Sinica，2021，8（9）：1523-1538.
14. [14] Tan Jiansuo, Wu Xinghua, Xu Wenguang, etc Industrial prediction model based on tSNE LSTM algorithm [J]. Modern Electronic Technology, 2024, 47 (12): 81-85

Funding

Key Technology Research Project for Intelligent Operation and Maintenance of Rail Transit (20090503100); International Joint Laboratory of China Laos Railway Engineering under the the Belt and Road Initiative (21210750300)

About the authors

Yu Fengqin (1999-), female, Han ethnicity, from Anqing, Anhui Province, is currently pursuing a master's degree in Intelligent Control and Monitoring of Rail Transit.