On-Demand Ride Services Demand Forecasting And Dynamic Pricing Model

As an emerging urban transportation mode,on-demand ride services have a significant impact on travelers’ mode choice behavior and even reshape urban transportation systems and human mobility.Compared with the traditional travel mode,on-demand ride services are scheduled and operated by an on-demand ride services platform,which is more convenient for real-time regulation and management.In this process,it is crucial to understand the behavior of on-demand ride services users and the regulation’s impact on the supply and demand.Based on the real-world datasets,this paper proposes a deep learning framework and a dynamic pricing optimization model,which can assist the on-demand ride services platform and government in regulating and managing the on-demand ride services market.The main research contents and results are reflected as follows:(1)Reveal the travel behavior mechanism:we analyze the on-demand ride services order data and reveal the temporal and spatial distribution patterns of different ride-sourcing services.Combined with the questionnaire data about ride-sharing travel behavior,we analysis ride-sharing users’ socio-demographic characteristics and quantify ride-sharing’s influence on vehicle use and purchase willingness based on a case study in Hangzhou,China.(2)Efficient demand forecasting:we analyze the spatial dependences,temporal dependences,and exogenous dependences among different influencing factors of the ride-sourcing passenger demand.Based on that,we propose a Deep Learning(DL)approach,named the Fusion Convolutional Long Short-Term Memory Network(FCL-Net)to address the temporal and spatial correlations,and use real-world datasets to calibrate and test the proposed model.(3)Dynamic pricing:we propose a dynamic vacant car-passenger meeting model to characterize the influence of short-term variances and disturbances of instant demand and supply in a ride-sourcing market.Based on the meeting model,we introduce the on-demand ride services’dynamic pricing optimization model and the corresponding Approximate Dynamic Programming(ADP)algorithm to solve the timing decision problem,and finally tests different dynamic pricing strategies under different targets.