A train delays prediction model under different causes based on MTGNN Approach

Train delays have an important influence on the daily train dispatching. This study proposes a deep learning model that combines graph learning, graph convolution and temporal convolution modules to predict train arrival delays under different causes. First, the multi-dimensional delay data onto different causes are obtained by data processing. Second, our model automatically extracts the spatial and temporal relationship between trains and railway infrastructural factors, through a graph learning module. Third, residual connections, skip connections and output model are combined to get more precise results. Then, a case study of the Wuhan-Guangzhou high-speed railway is illustrated to verify the performance of the proposed model for different delay causes. Quantifying the impact of infrastructural factors on the prediction results of the model under different causes, the results show that the number of tracks improves the accuracy of model prediction more than the length of the section. Comparing different kinds of candidate models, the results show that the proposed model can well deal with the spatial-temporal relationship of delayed trains and is superior to other baseline models (1.e., Back Propagation, Gate Recurrent Unit, and Temporal Graph Convolutional Network).