Research On Vehicle Resource Scheduling And Trading Algorithm Based On Deep Reinforcement Learning

The shared vehicle riding market has developed rapidly in recent years,playing an important role in allocating urban transportation resources and promoting sharing and low-carbon economic development.Consequently,the related problems of vehicle resource scheduling optimization and operators’ dynamic pricing strategy have attracted extensive attention from industry and academia.With the development and application of artificial intelligence（AI）technology,numerous methods based on deep reinforcement learning techniques have been proposed to solve vehicle resource scheduling by optimizing the distribution of vehicle resources in the spatial dimension to balance the supply-demand relationship.However,the rising demand for riding services and the total amount of vehicle resources in the shared vehicle riding system are changing dynamically in space and time.There is a need for an effective resource management mechanism to optimize the distribution of vehicle resources in the time dimension.Therefore,based on deep reinforcement learning technology,this thesis proposes a cruise vehicle dispatching and resource management（CVDRM）method with an adaptive reward mechanism and multi-objective optimization.With supply-demand relationships and contextual information such as weather,time,and geographic location,the CVDRM method can pre-schedule cruising vehicles to a heavy-load grid area to balance the supply and demand in the spatial dimension.With the relative relationship of request success rate and vehicle matching rate,the CVDRM method adaptively adjust vehicle resources in the shared ride system to balance the supply and demand in the time dimension.Simulation results show that the CVDRM method effectively improves operator’s income while balances the request success rate and vehicle resource utilization at acceptable levels.At the same time,the empty running time of shared taxis and the waiting time of users is shortened.Also,the development of the shared vehicle riding market has gradually become a situation of multi-operator competition and coexistence.However,the research on vehicle resource trading and dynamic pricing strategies is still limited to single-operator scenarios.In order to cope with the challenges brought by t the increased freedom of user choice and multi-operator competition and coexistence,it is necessary to balance the supplydemand relationship between users and the competitive relationship between operators and operators,as a result,promote the sustainable and healthy development of the shared vehicle riding market.Therefore,the thesis establishes a market framework for vehicle resource trading in multi-operator scenarios based on blockchain technology.Furthermore,to maximize the operator’s income and improve the user’s interests under the market framework,the thesis proposes a competitive pricing scheme（CPS）under dynamic pricing based on supply-demand and market competition.Based on noncooperative game theory,the vehicle resource trading between multiple operators and users is modeled as a two-stage Stackelberg game.Finally,the multi-agent reinforcement learning algorithm is proposed to solve the nash equilibrium（NE）to obtain the optimal resource trading strategy.The simulation results show that the pricing scheme proposed can effectively improve the operator’s profit,and then motivate operators to participate in vehicle resource trading under the market framework.In addition,the competitive pricing scheme improves the user benefits while increasing the operator’s profit,thereby enhancing the user’s purchasing intension and the operator’s market attractiveness.