Research On Demand Prediction And Dynamic Pricing Of Car-sharing Based On Big Data

The car-sharing especially the electric car rental service is rapidly developed and serves as a important transportation mode in recent years.Car-sharing rental mode can not only satisfy the diversified and personalized demands of travelers but also improve the resource utilization,which effectively alleviates the problem of traffic congestion,air pollution and resource waste.In order to promote the healthy and sustainable development of car-sharing industry,some reasonable operation strategies should be established to improve the economic benefits.However,due to the randomness of users’ travel,the car-sharing operators are generally faced with the problem of unbalanced supply and demand of vehicles at the car rental stations.What’s more,the high investment with low profit also disturbs the operators of car-sharing.Therefore,to guarantee the safe,stable and efficient operation of the car-sharing industry,it is necessary to scientifically predict users’ travel demand,make a reasonable dynamic pricing strategy to adjust the balance between supply and demand and consider how to improve operating profit.Based on a large car-sharing operation system in China,the pricing problem of each site in different periods is studied in this paper to provide a scientific dynamic pricing strategy for car-sharing operators.Firstly,based on the actual operation data of car-sharing,the macroscopic and microscopic travel characteristics of users and the law of supply and demand of the sites are analyzed,the time period and sites are classified.Secondly,the model of time series based on statistics and the model of GBDT based on machine learning,are designed to predict the demands of users in different periods and different sites.The prediction accuracy of the two models is compared.Thirdly,based on the classification and prediction results,considering the business model,operating costs,user acceptance,elasticity of demand price and other factors,the dynamic pricing model is established to maximize the operator’s profit.The constraint conditions of the model are determined.The relationship between price and demand and scheduling cost is clarified.Also,the immune genetic algorithm is designed to solve the dynamic pricing model.Finally,the practicability and effectiveness of the research in this paper are verified by using the actual operation data of the car-sharing enterprises.The results show that the demand prediction model based on GBDT has higher precision and can provide better data support,which makes up for the shortage of time series model.Also,the dynamic pricing model based on the real-time demand can not only effectively reduce the imbalance of the shared vehicles,but also optimize the practical problems such as reducing the scheduling cost and increasing the order quantity and profit.