Research On Dynamic Dispatching And Pricing Method Of Connected Vehicles

In recent years,with the development of shared mobility,connected vehicles have played an increasingly important role in relieving traffic congestion and providing travel convenience.However,how to achieve a balance between supply and demand on the connected-vehicle platform is an important issue,given the real-time travel demand in large-scale transportation networks and the randomized location of vehicles.Existing research has addressed the problem mainly at the grid level through vehicle dispatching methods,but there is still room for exploration of cooperative vehicle dispatching for real roads or jointing pricing.In this paper,we focus on the dynamic vehicle dispatching and pricing problem of the connectedvehicle platform under different traffic network scales.The specific research contents and contributions are as follows.First,an actor-critic algorithm based on causal exploration is proposed for the problem of cooperative dispatching among large-scale vehicles.The proposed algorithm uses feed-forward neural networks to design three modules:causal network,actor network,and centralized critic network.During the exploitation process,the optimal dispatching policy is generated based on the actor network.During the exploration process,the causal network is constructed to measure the influence of neighborhood’s decisions on the current region,and vehicles are dispatched according to causal impact weights.In addition,the merits of the executed actions are evaluated based on the critic network.The experimental results show that the proposed algorithm has significant improvements in order response rate,cumulative driver income and driver revenue efficiency.Second,considering the correlation between pricing and dispatching in the connected-vehicle platforms,a combined multi-agent reinforcement learning and network flow algorithm is proposed to achieve joint optimization of pricing and dispatching.Specifically,based on the multiagent framework,zone is abstracted as agent,and each zone generates a spatial-temporal pricing strategy by capturing the dynamic supply and demand changes of its neighbors.Then,considering the driver’s expected future profit of drivers and the supply and demand distribution,the vehicle dispatching problem is transformed into a network flow optimization problem,with the objective of maximizing the total net profit of drivers to achieve coordinated dispatching.The simulation results show that the proposed algorithm can improve the satisfaction of the platform,the driver and the passenger.Third,a deep reinforcement learning algorithm based on graph convolution networks is designed to optimize vehicle dispatching strategy to address the supply-demand imbalance problem at the road level.The previous study modelled urban roads at the grid level but ignored the physical network topology.This paper further utilizes a graphical representation of the road network to better reflect the underlying geometric structure.The proposed algorithm uses the graph convolutional network to estimate the action value function of the all roads,and efficiently dispatches vehicles to the appropriate roads based on the softmax-action selection strategy.Finally,a simulator reflecting the real road network is designed as a training and testing environment,and the effectiveness of the proposed algorithm is verified.