Short-term urban traffic flow prediction using deep spatio-temporal residual networks

Traffic flow prediction, as the key technology of traffic guidance system (TGS), is of great importance to mitigate traffic congestion and city management. In view of the existing research mainly considering the adjacent area and ignoring the influence of far area on the current section, this paper presents a method, called ST-ResNet, for predicting urban traffic flow the predicts all roads in an area, which takes into account not only the time information but also the effects of nearby and beyond areas. We use two basic traffic parameters, volume and speed, as the input of the model to simultaneously predict the traffic volumes and average speed of the road segment. Experiment on all motorways and `A' roads managed by the Highways Agency, known as the Strategic Road Network (SRN), in England demonstrate its great and precise accuracy of short-term traffic flow prediction.