Taxi Dispatch Under Different Willingness of Sharing: A Reinforcement Learning Approach

With the improvement of the living standards of urban residents, demand-responsive public transport, such as taxis has gradually become one of the main choices for people to travel, and the addition of ride-sharing also provides a new solution to the imbalance between taxi supply and demand as well as urban traffic congestion. However, the existing ride-sharing taxi dispatching methods do not give enough consideration to passengers’ ride-sharing willingness. Therefore, a taxi dispatching method considering ride-sharing behavior based on reinforcement learning is proposed in this study. First, the LightGBM model, which is used to dispatch taxis ahead of time to avoid hysteresis, was utilized to forecast taxi demand. Second, the state, action, and reward were modeled with taxi demand and vehicle data in the research area based on the hexagonal grid matrix; Finally, considering the different passengers’ ride-sharing willingness, a ride-sharing model was established based on the taxi dispatching model. The proposed model is validated through a simulation experiment. The results showed that the dispatching model could obtain satisfactory results under different dispatching objectives, demand pressures, and willingness to ride-sharing on both the average passenger waiting time and average taxi dispatching distance.