Urban Vehicle Trip Chain Reconstruction Based on Gradient Boosting Decision Tree

A reconstruction method for urban vehicle trip chain based on gradient boosting decision tree was proposed to extract actual vehicle trajectories for transportation planning, traffic design, management and evaluation. Firstly, vehicles were matched by license plate number (LPN), and the corresponding travel chains sorted by time stamp were initially extracted and split according to the intersection adjacency matrix and estimated link travel time. Subsequently, the key features that affecting vehicle route choice were identified based on travel behavior analysis and traffic conditions, and a reconstruction method for local lost trip chain was developed based on gradient boosting decision tree (GBDT). Finally, taking the field LPN data from Nanming district of a Chinese city as an example, the accuracy and calculation efficiency of the proposed method and existing ones were verified. The result shows that the proposed method can achieve a high reconstruction accuracy of 91%, and it superior to the traditional ones in urban vehicle trip chain reconstruction. © 2020, Editorial Department, Journal of South China University of Technology. All right reserved.