Hybrid Approach to Train Delay Prediction: An Integration of Analytical Model and Deep Learning Techniques

Abstract

High-speed rail (HSR) operation has the distinctive features of rapid speed and high density. When external environment fluctuations or system malfunctions interfere with the normal operation of certain trains, it may lead to train delays or even cascading delays, substantially compromising the normal functioning of the HSR system and adversely affecting passengers’ travelling experience. To tackle this, this article proposes a hybrid train delay prediction model, which can effectively aid dispatchers in making optimized dispatching strategies and help passengers in replanning their trips. Taking into account of analytical mode, we identify the factors that influence train running and divide them into three categories. Then, a transformer-based network structure is constructed for predicting the delay time of the target train at the target station. Moreover, a loss function is designed in according to the root mean square error (RMSE) and the delay variation rules. Subsequently, numerous experiments are conducted based on the actual operation data of Chinese HSR subnetwork, and the results illustrate that our proposed model outperforms the baseline model substantially, with the RMSE and the mean absolute error (MAE) improved by 18.39% and 12.16% at most respectively, thereby verifying the superiority of our proposed model.