Driver Back-Tracing Based on Automated Vehicle Identification Data

Traffic congestion causes traveler delay, environmental deterioration, and economic loss. Most studies on congestion mitigation focus on attracting travelers to public transportation and expanding road capacity. Few studies have been found to analyze the contribution of different traffic flows to the congestion on roads of interest. This study proposes an approach to driver back-tracing on the basis of automated vehicle identification (AVI) data for congestion mitigation. Driver back-tracing (DBT) aims to estimate the sources of the vehicles on roads of interest in both spatial and temporal dimensions. The spatial DBT model identifies the origins of vehicles on the roads and the temporal DBT model estimates the travel time from the origins to the roads. The difficulty lies in that vehicle trajectories are incomplete because of the low coverage of AVI detectors. Deep neural network classification and regression are applied to the spatial and temporal DBT models, respectively. Simulation data from VISSIM are collected as the dataset because of the lack of field data. Numerical studies validate the promising application and advantages of deep neural networks for the DBT problems. Sensitivity analyses show that the proposed models are robust to traffic volumes. However, turning ratios, and the number and layout of AVI detectors may have noticeable impacts on the model performance.