DTFPRS: A Distributed Traffic Flow Prediction Framework Based on Road Segmentation

Qiqi Cai, Jian Cao, Shiyou Qian, Yirong Chen

Abstract: Accurate traffic flow prediction is of great significance. Recently, lots of deep neural network models have been applied in predicting traffic flow, such as Attention based Spatial-Temporal Graph Neural Network(ASTGNN). However, although these models have good prediction performances, there are still some problems when they are applied to the increasingly expanding and complex traffic network, such as memory consumption and computing overheads. Therefore, this paper proposes a Distributed Traffic Flow Prediction Framework Based on Road Segmentation(DTFPRS). Specifically, it is consisted of Spatial Partitioning (SP) and Traffic Flow Prediction(TFP). SP segments the road into sub-roads according to the correlation of traffic flow, and TFP trains ASTGNN on each sub-road. Experiments on real-world traffic flow datasets demonstrate that DTFPRS can greatly shorten the training time and reduce the size of the model without affecting the accuracy of the prediction results which makes DTFPRS can be used to large traffic networks.