Multiagent Reinforcement Learning-Based Taxi Predispatching Model to Balance Taxi Supply and Demand

Abstract

With the improvement of people’s living standards, people’s demand of traveling by taxi is increasing, but the taxi service system is not perfect yet; taxi drivers usually rely on their operational experience or cruise randomly to find passengers. Without macroguidance, the role of the taxi system cannot be fully utilized. Many scholars have studied taxi behaviors to find better operational strategies for drivers, but their researches rely on local optimization methods to improve the profit of drivers, which will lead to imbalance between supply and demand in the city. To solve this problem, we propose a Multiagent Reinforcement Learning- (MARL-) based taxi predispatching model through analyzing the running data of 13,000 taxis. Different from other methods of scheduling taxis based on the real-time location of orders, our model first predicts the demand for taxis in different regions in the next period and then dispatches taxis in advance to meet the future requirement; thus, the number of taxis needed and available in different regions can be balanced. Besides, in order to reduce computational complexity, we propose several methods to reduce the state space and action space of reinforcement learning. Finally, we compare our method with another taxi dispatching method, and the results show that the proposed method has a significant improvement in vehicle utilization rate and passenger demand satisfaction rate.