DC-ST: A Short-term traffic flow prediction Approach based on Distance Correlation and Spatial-Temporal Dependence

Accurate traffic flow prediction are very important for route planning, traffic control and intelligent driving. The uncertainty and complexity of traffic flow states makes it a challenging problem. In this work, we proposed a distance correlation based spatial-temporal prediction model. The double filtering criterion is developed to ensure the spatial-temporal validity and reduces the redundancy. Then, the analysis object is refined from spatial nodes to spatial-temporal segments by dynamically slicing techniques, which leads to a more in-depth mining in temporal dimension. Finally, the effectiveness of the our proposed model are tested by experiments on real traffic flow datasets.