High-Resolution Traffic Flow Prediction Model Based on Deep Learning

Abstract

The traditional platoon dispersion model is based on the hypothesis of probability distribution, and the time resolution of the existing traffic flow prediction model is too big to be applied to the adaptive signal timing optimization. Based on the view of the platoon dispersion model, the relationship between vehicle arrival at downstream intersection and vehicle departure from the upstream intersection was analyzed. Then, the high-resolution traffic flow prediction model based on deep learning was proposed. The departure flow rate at the upstream was taking as the input and the arrival flow rate at downstream intersection was taking as the output in this model. Finally, the parameters of the proposed model were trained by the field survey data, and this model was implemented to predict the arrival flow rate of the downstream intersection. The result shows that the proposed model can better reflect the fluctuant characteristics of traffic flow and reduced the sum of the squared errors (SSE), MSE, and MAE by 13.17%, 13.21%, and 14.24%, compared with Robertson’s model. Thus, the proposed model can be applied for real-time adaptive signal timing optimization.