Fine-grained predicting urban crowd flows with adaptive spatio-temporal graph convolutional network

Abstract

Predicting crowd flows is important for traffic management and public safety, which is very challenging as it is affected by many complex factors. In this paper, we propose a novel fine-grained predicting urban crowd flows approach with an adaptive spatio-temporal graph convolutional network, called ASTGCN. This approach can simultaneously predict the inflow, outflow, and flow direction. We first design a method for modeling crowd flow in irregular urban regions based on urban bus line data. Then, we design an end-to-end structure of the adaptive spatio-temporal graph convolutional network with unique properties of spatio-temporal data. Finally, extensive experiments on GAIA open dataset are constructed to evaluate the performance of ASTGCN. Results show that our approach outperforms four well-known methods, the average absolute error is reduced by 28.7 %, and the root mean square error is reduced by 37.9 %.