Ride-Sharing Matching of Commuting Private Car Using Reinforcement Learning

Junchao Lv, Linjiang Zheng, Longquan Liao, Xin Chen

Abstract: Commuting private car is one of the main factors causing traffic congestion and low traffic efficiency in the morning and evening peak, and the ride-sharing matching of commuting private car is an effective measure to solve traffic problems. However, limited by the difficulty of obtaining private car trajectory data, current research in the field of transportation rarely involves private car. Electronic Registration Identification (ERI) of motor vehicles technology brings hope for the development of related research. This paper proposes a set of methods for commuting private car ride-sharing matching based on ERI data. The main research content includes two parts: commuting private car identification and commuting private car ride-sharing matching. In the commuting private car identification, information entropy and coefficient of variation are used to characterize the regularity of travel, and private cars with regular travel heights are identified as commuting private cars through random forest classification. In the commuting private car ride-sharing matching, reinforcement learning is combined with the ride-sharing matching scenes to realize the matching from a global and more foresight perspective instead of focusing on even satisfaction. Finally, we apply the ERI data of Chongqing to proposed method. The experimental results show that the method can accurately identify commuting private car, and the number of commuting private car is reduced by 21.45% and 21.01% respectively in the morning and evening peak after ride-sharing matching.