This is the official implementation of the following paper:

Topological Graph Convolutional Network-Based Urban Traffic Flow and Density Prediction

Han Qiu, Qinkai Zheng, Mounira Msahli, Meikang Qiu, Gerard Memmi, Jialiang Lu

Abstract: With the development of modern Intelligent Transportation System (ITS), reliable and efficient transportation information sharing becomes more and more important. Although there are promising wireless communication schemes such as Vehicle-to-Everything (V2X) communication standards, information sharing in ITS still faces challenges such as the V2X communication overload when a large number of vehicles suddenly appeared in one area. This flash crowd situation is mainly due to the uncertainty of traffic especially in the urban areas during traffic rush hours and will significantly increase the V2X communication latency. In order to solve such flash crowd issues, we propose a novel system that can accurately predict the traffic flow and density in the urban area that can be used to avoid the V2X communication flash crowd situation. By combining the existing grid-based and graph-based traffic flow prediction methods, we use a Topological Graph Convolutional Network (ToGCN) followed with a Sequence-tosequence (Seq2Seq) framework to predict future traffic flow and density with temporal correlations. The experimentation on a real-world taxi trajectory traffic data set is performed and the evaluation results prove the effectiveness of our method.

If you have any question, please raise an issue or contact qinkai.zheng1028@gmail.com.

Datasets can be downloaded via the link.

• numpy==1.19.1
• torch==1.3.0
• torchsummary==1.5.1
• tensorboardX==2.1