ABSTRACT
The estimation of train delays is important for making timetables, dispatching trains, and planning infrastructures. The prediction of train delays, however, mainly depends on metro traffic controllers’ experience up to the present. Focusing on the recognition and prediction of train delays, this paper first analyzes the relationship between station passenger flow, schedule train working diagram, and actual train working diagram. The neural network is then introduced to recognize and predict the train delays. Based on classical CNN (convolutional neural network) and AlexNet network, a train working diagram recognition model with five neural layers and three pooling layers is developed. The proposed model is constructed in the framework of TensorFlow. At last, the effectiveness of the proposed model is evaluated using a real-world data set. The prediction results are consistent with the actual train working diagram and have an accuracy of about 80%. The method is expected to assist metro dispatchers to deal with train delays and to improve the overall operation service level of rail transit.
