Efficient Dispatching for On-Demand Ride Services: Systematic Optimization Via Monte-Carlo Tree Search
Xiqun (Michael) Chen,Xiaowei Chen,Hongyu Zheng,Feng Xiao
DOI: https://doi.org/10.1016/j.trc.2021.103156
2021-01-01
Abstract:Efficient dispatching for on-demand ride services is essential to the ride-sourcing platform, passengers, and drivers in a competitive ride-sourcing market. The existing first-come-first-serve (FCFS) dispatching method is a myopic mechanism that sought to minimize individual-level waiting time. To improve the passenger-vehicle matching efficiency and reduce the cancelation rate, a searching tree structure is developed to deal with the vehicle dispatching problem considering vehicles still in service. This paper formulates a distinctive systematic dispatching model from the passengers' perspective and presents new dispatching rules to match passengers and vehicles. Based on an improved tree policy and efficient branch reduction policy, we customize and enhance the reinforcement learning approach, Monte-Carlo Tree Search (MCTS), to solve the multi-period sequential dispatching problem and enable efficient dispatches. The computational Complexity of MCTS is analytically derived. As benchmarks, the Hungarian algorithm, CPLEX, greedy algorithm, and two global optimization algorithms are employed. We compare these algorithms by using both simulation and real-world city-scale on-demand ridesourcing data. Both numerical and city-scale experiment results show that the improved MCTS dispatching approach increases the percentages of satisfied passengers and dispatched drivers, and reduces the number of unmatched passengers and cumulative waiting time compared to the benchmark algorithms. The results shed light on modeling urban dispatching problems with available vehicles in both forward-looking and backward-looking time periods. This paper demonstrates that improved MCTS performs well in solving a multi-period sequential optimization problem in real-world city-scale applications.