Short-time passenger flow prediction of new urban rail transit based on graph convolutional neural network

Abstract

Short-term passenger flow forecast is an important task in urban rail transit operation. Emerging deep learning technologies are seen as an effective way to solve this problem. In this study, we propose a deep learning model called GCN-Conv2d, which combines graph convolutional networks and two-dimensional convolutional neural networks. Firstly, a GCN model is introduced to deal with three passenger flow patterns (recent, daily, and symmetrical). The GCN model can extract the spatio-temporal correlation and topological information within the entire network, and then the two-dimensional convolutional neural network can be applied to deeply integrate the passenger flow information, and the latter can extract the spatio-temporal features between different passenger flow patterns and between stations at different distances. Finally, a fully connected layer is used to output the results. The GCN-Conv2d model predicted the smart card data of Zhuzhou Intelligent rail express system at a time interval of 10 minutes. The results show that the error of the model in RMSE and MAE is smaller than that of the random forest model and the CNN model, which shows good performance. This study can provide important support for public transport operators to optimize the operation of urban rail transit and promote the intelligent operation of urban rail transit network.