AF-TCP: Traffic Congestion Prediction at Arbitrary Road Segment and Flexible Future Time

Xuefeng Xie, Jie Zhao, Chao Chen, Lin Wang

Abstract: Traffic congestion prediction is a fundamental yet challenging problem in Intelligent Transportation Systems (ITS). Due to the large scale of the road network, the high nonlinearity and complexity of traffic data, few methods are well-suited for citywide traffic congestion prediction. In this paper, we propose a novel deep model named AF-TCP for Traffic Congestion Prediction at Arbitrary road segment and Flexible future time. For the model to achieve the prediction at flexible future time, we construct all time representations in a unified vector space and further improve the model’s perception ability to different horizons. On the other hand, to realize the congestion prediction of arbitrary road segment within the city, we utilize road attributes and local neighbor structure to build the road segment representation, and design a deep model to fuse it with the corresponding historical traffic data. Extensive experiments on the real-world dataset demonstrate that our model exhibits stable performance at different prediction horizons and outperforms the baselines.