Deep reinforcement learning for urban multi-taxis cruising strategy

Taxis play an important role in urban transportation system. Efficient taxi cruising strategies are helpful to alleviate urban traffic congestions, reduce pollution emission, attenuate greenhouse gas, and also provide a fast service for passengers. However, in real scenarios, taxis cruising strategies are mostly based on their own experiences. At unfamiliar urban areas or during off-peak hours, drivers usually have no good idea for an optimal cruising strategy, which causes a low efficiency for service and also increases taxi operation costs. Considering that it is difficult to construct an analytical model for the taxis scheduling and cruising, in this paper, we put forward a data-driven model for multi-taxis cruising based on reinforcement learning. Furthermore, an evolutionary reinforcement learning method is proposed, which aims at improving the exploration of reinforcement learning and enhancing reinforcement learning to maximize the global reward in multi-agent tasks. In the experimental part, two other kinds of deep Q-learning methods and a roaming strategy are employed in the comparisons. The results demonstrate the superiority of our proposed algorithm.