Good or Mediocre? A Deep Reinforcement Learning Approach for Taxi Revenue Efficiency Optimization

摘要

Recently, with the rapid expansion of cities, optimizing taxi driving routes for improving taxi revenue efficiency has become the core issue of taxi system. However, most current research focuses on increasing platform revenue instead of improving drivers' revenue in a centralized dispatch taxi system just like DiDi, which results in a slower driver income growth and greater difficulties for recruiting drivers. To solve this problem, we propose a strategy of deep reinforcement learning based on driver mode. Firstly, the sequence selection process of drivers is modeled as markov decision-making process in driver mode. Then, we propose a learning scheme based on deep Q network to optimize the driver's decision-making strategy. We know that the real selection of historical taxi drivers is very helpful to the selection of current taxi drivers, so we choose the historical record of the current location as the edge data to update the edge network. Finally, we used a real data set generated by more than 1,400 taxis in Changsha. The simulation experiments show that our scheme reduced cruising time of taxis and improved the driver's income by 4-5%. The carbon emissions are obviously reduced by saving almost 6% fuel consumption, which contributes significantly to green mobility.