Train Arrival Delay Prediction Based on a CNN-LSTM Approach

Abstract

Train delay prediction can assist in the reasonable formulation of a train diagram, and provide the basis for the decision-making of train dispatchers. In this paper, a hybrid method combining convolution neural network (CNN) and long short-term memory network (LSTM) is proposed to predict train arrival delays. First, eight characteristics (e.g., train departure delay, train actual running time) affecting train arrival delay are selected as the initial input variables of the prediction model. Next, CNN extracts feature again based on eight features, and outputs 32 new features. Further, combined with the newly extracted features, the proposed prediction model is trained using LSTM. Finally, the prediction performance of the proposed CNN-LSTM prediction model is evaluated based on the real-world operation records of the Wuhan-Guangzhou high-speed railway. The case study results show that the proposed prediction model has a higher prediction accuracy and is better than deep neural networks and LSTM.