Optimal Transportation Network Company Vehicle Dispatching via Deep Deterministic Policy Gradient

With the popularity of smart phones and the maturity of civilian global positioning system (GPS) technology, transportation network company (TNC) services have become a prominent commute mode in many major cities, which can effectively pair the passengers with the TNC vehicles/drivers through mobile applications. However, given the growing number of TNC vehicles, how to efficiently dispatch TNC vehicles poses crucial challenges. In this paper, we propose a novel method for TNC vehicle dispatching in different areas of the city based on deep reinforcement learning (DRL) method with joint consideration of the TNC company, individual TNC vehicle, and customer/passenger. The proposed model optimizes the distribution of vehicles geographically to meet the customers’ demands, while improving the drivers’ profit. In particular, we consider the high dimensional state and action space in the urban city traffic dynamic environment, and develop a deep deterministic policy gradient, an actor-critic based DRL algorithm for dispatching vacant TNC vehicles. We leverage Didi Chuxing’s open data set to evaluate the performance of the proposed approach, and the simulation results show that the proposed approach improves the average income of the driver while satisfying the supply and demand relationship between TNC vehicles and customers/passengers.