Gaussian Noise Data Augmentation-Based Delay Prediction for High-Speed Railways

Abstract

Accurately predicting delays for high-speed railways (HSRs) is a challenging yet significant task. The historical operation data of the HSRs, implicating delay derivation rules under the dispatchers’ rescheduling strategies, have sparsity characteristics, resulting in heterogeneous prediction performances under different scenarios. This article proposes a Gaussian noise data augmentation-based delay prediction method to cope with the sparsity. Specifically, the Gaussian noise is added to the original data based on the train operation data characteristics. Then, the delay data rather than the full-state dataset are selected as the training data for different designed machine learning prediction models. Numerous studies based on real HSR operational data from the Beijing Railway Bureau show that the proposed method could significantly improve the prediction accuracy under different scenarios with different machine learning models, verifying the effectiveness of the performance improvement. The relevant results could be significantly helpful for real-time train rescheduling and passenger management, thus improving the emergency response capabilities of HSRs.