Fleet Optimization of Smart Electric Motorcycle System Using Deep Reinforcement Learning

Abstract

Smart electric motorcycle-sharing systems based on the digital platform are one of the public transportations that we use in daily lives when the sharing economy is considered. This transportation provides convenience for users with low-cost systems while it also promotes an environmental conservation. Normally, users rent the vehicle to travel from the origin station to another station near their destination with a one-way trip in which the demand of renting and returning at each station is different. This leads to unbalanced vehicle rental systems. To avoid the full or empty inventory, the electric motorcycle-sharing rebalancing with the fleet optimization is employed to deliver the user experience and increase rental opportunities. In this paper, the authors propose a fleet optimization to manage the appropriate number of vehicles in each station by considering the cost of moving tasks and the rental opportunity to increase business return. Although the increasing number of service stations results in a large action space, the proposed routing algorithm is able filter the size of the action space to enable computing tasks. In this paper, a Deep Reinforcement Learning (DRL) creates the decision-making function to decide the appropriate action for fleet allocation from the last state of the number of vehicles at each station in the real environment at Suranaree University of Technology (SUT), Thailand. The obtained results indicate that the proposed concept can reduce the Operating Expenditure (OPEX).