Research On Allocation And Configuration Strategy Of Reserved Parking Resources

In recent years,as China’s urbanization and motorization levels continue to rise,parking difficulties have become a major problem hindering high-quality urban development.With the emergence of new technologies such as 5G communication,Internet of Things and cloud computing,a new model of parking reservation has emerged,providing new ideas to solve the problem of parking difficulties.However,at the present stage,reservation parking fails to take into account the heterogeneity of multiple parking lots and the dynamic nature of demand.Therefore,based on the research of "cloud-net-end integrated city-level intelligent parking service technology"(2019YFB1600304),this paper constructed a multi-objective dynamic weight time sharing reservation model for multi heterogeneous parking lot scenarios,adopted dynamic adjustment strategies,and achieved dynamic allocation of reserved parking resources.In addition,for some randomly scheduled travel scenarios,this paper constructed a reinforcement learning model for real-time reservation resource allocation to achieve dynamic allocation of reserved parking resources from a long-term perspective.First,this paper constructed a time-share reservation model for multiple parking lots in a neighborhood with different parking capacities,parking rates,walking distances and user preferences,with the objectives of maximizing the parking lot utilization time and minimizing the parking cost of travelers.The model’s target weights were dynamically set by combining the high and low peak characteristics of parking demand,and the adjustment strategy of reallocating the reservation requests that have not yet arrived at the parking lot was designed based on the characteristic that reservations can be made in advance for multiple time periods.The branch-and-bound method was used to solve the model.Through multiple model comparisons,the results show that the proposed adjustment strategy can significantly improve the average utilization rate and the acceptance rate of reservation requests in the reservation area;the dynamic weights can better adapt to the high and low peak variation of demand;and the dynamic pricing can effectively balance the resource utilization of each parking lot and distribute the demand in a balanced manner.Secondly,this paper constructed a real-time reservation parking resource optimization configuration model based on reinforcement learning for a real-time reservation parking scenario where reservation parking and non-reservation parking coexist,with the current supply and demand state and future supply and demand forecast as the state function,the opposite of the average travel time of all parkers as the reward function,and the reservation space allocation ratio of each parking lot for each time period as the action function.The REINFORCE(RF)algorithm,Advantage Actor-Critic(A2C)algorithm and Proximal Policy Optimization(PPO)algorithm were used to solve the scenarios with five different reservation demand ratios and compared them with the first-come-first-served model and the short-sighted model.The results show that the model proposed in this paper can effectively save travelers’ travel time,and the travel time gradually decreases as the reservation demand ratio increases;when the reservation demand ratio is low,the model tends to allocate unpopular parking resources to reserved vehicles,while as the reservation demand ratio increases,the model reserves popular parking resources for reserved vehicles and makes non-reserved vehicles go to unpopular parking lots to park.