Road section traffic flow prediction method based on the traffic factor state network

Large-scale and diversified traffic data resources strongly support research into estimating urban traffic states and predicting traffic flow. There are many studies on traffic prediction, but there is still not a universally applicable real-world traffic flow prediction method. This paper regards urban road sections as a microscopic traffic system. Based on a deep understanding of the traffic state of road sections, it proposes a pertinent traffic flow prediction framework based on the traffic factor state network (TFSN) framework by combining model-driven methods with machine learning to identify traffic patterns in road sections. For different road traffic patterns, it proves mathematically that the state of traffic flow in each period tends to be the state of the corresponding period with greater probability. According to different road patterns and traffic states, suitable traffic flow modeling and prediction methods were selected. The case shows that this method can improve the accuracy of traffic flow predictions. The research results demonstrate that the average absolute percentage error of traffic flow predictions in urban sections selected with different characteristics and models is reduced by 7.51% compared with the direct prediction error method, verifying the effectiveness and usability of the proposed prediction framework.