Multi-source Data-driven Dynamic Pricing Model Of On-street Parking Charging Strategy

The integrated innovation of the new generation information technologies,such as big data and artificial intelligence,has promoted the digitalized,networked and intelligent transformation of parking fields.To meet the demand for dynamic management and control of intelligent parking,this study takes on-street parking facilities as the research object and proposes parking charging strategies under the mobile Internet environment,while analyzing on-street parking choice behaviors under the influence of different strategies.Based on the prediction of parking occupancy,a dynamic pricing model for on-street parking charge is created.This research integrates data such as questionnaires,berth utilization information,weather and point of interest.This model can effectively deliver on the real-time,initiative and refined management of on-street parking.Firstly,two on-street parking charging strategies,i.e.,the reward-based and the differentiated charging,are proposed for intelligent parking.A questionnaire survey on on-street parking choice behavior under the influence of different strategies is carried out,and multinomial Logit choice behavior models for on-street parking under the two charging strategies are constructed respectively.And the influencing factors for on-street parking choice behavior are analyzed,including personal attribute,weather,parking frequency,surrounding roads’ conditions,parking occupancy,reward amount,parking fees and others.It is shown that parking fee,reward amount,parking occupancy and surrounding roads’ conditions would deliver relatively significant influence on on-street parking choice behavior.Secondly,in combination with multi-source data,such as real-time information on on-street parking,weather and points of interest,this study proposes a short-term prediction method for parking occupancy with spatiotemporal graph convolution network based on deep learning.By integrating Long Short-Term Memory Network(LSTM)and Graph Convolution Neural Network(GCN),this method can obtain the spatiotemporal characteristics of parking occupancy.In addition,Residual Network(Res Net)is adopted to address the problem of gradient explosion or disappearance in deep learning.The prediction accuracy of model proposed in this paper is higher than other models.On the basis of the above research,combined with drivers’ on-street parking choice behavior and the result of parking occupancy prediction,this study comprehensively considers the balanced utilization of the on-street parking spatiotemporal resources and the economic benefits of parking facilities’ operators in the mobile Internet environment from the perspectives of system managers and such operators.Dynamic pricing models are constructed separately for different charging strategies,time windows and on-street parking facilities.These models take the minimization of the difference index of the total regional occupancy and the maximization of economic benefits as their optimization objectives,while taking price fluctuation control and parking space capacity limitation as constraints.And a simulated annealing genetic algorithm is designed to solve the models.Finally,certain on-street parking facilities in Yinzhou District,Ningbo,are selected for empirical research.The spatiotemporal distribution of on-street parking occupancy is uneven in this area,with many vehicles cruising around their destinations to seek a vacant parking lot.Based on the prediction results for real-time parking occupancy,the optimal dynamic pricing schemes both for reward-based and differentiated charging methods are formulated respectively.The results show that both schemes can strike an effective balance for the spatiotemporal resource utilization of on-street parking space in the region.The profit from the reward-based strategy is slightly higher than that from the differentiated charging method,but the latter is more helpful for reducing the parking occupancy in high-demand peak hours.