A delay prediction model for high-speed railway: an extreme learning machine tuned via particle swarm optimization

Abstract

Train delay prediction is a significant part of railway delay management, which is key to timetable optimization of Highspeed Railways (HSRs). In this paper, an extreme learning machine (ELM) tuned via particle swarm optimization (PSO) is proposed to predict train arrival delays of HSR lines. First, five characteristics (e.g., the plan running time between the present station and the next station, stations) are selected from nine characteristics as input variables for ELM by correlation coefficient matrix. Next, PSO algorithm is implemented to effectively resolve the hyperparameter adjustment of ELM, which overcomes tedious manual regulation for the number of hidden neurons. Finally, a case study of fifteen stations on Beijing-Kowloon (B-K) HSR line in China is proposed using the ELM tuned via PSO (ELM-PSO). The prediction performance of the proposed method is verified by comparison with six benchmark models. The results indicate that our method is superior to these baseline models in prediction accuracy.