Research On Demand Forecasting And Dynamic Pricing Method Of Ride-hailing Based On Large-scale Datasets

With the rapid development of online car-booking(e-hailing)platforms such as Didi and Uber,e-hailing has become an important part of the rental car market and an important supplement to urban public transport.Compared with cruise taxis,the on-demand dispatching mode of e-hailing improves the information matching ability between passenger demand and e-hailing supply.However,due to the uneven distribution of demand and supply in time and space,there will still be problems such as delayed response and poor supply and demand matching.In addition,the effect of price leverage as a favorable means to adjust the relationship between supply and demand of online car hailing has not been effectively played.When passenger demand is in a state of saturation or supersaturation,dynamic pricing(price increase)can increase the cost of passenger cars on the one hand,which is conducive to reducing part of oversaturated taxi demand and promoting supply and demand to reach a balanced state.On the other hand,it is beneficial to encourage car-hailing providers to actively increase the number of online drivers and vehicles to meet the excess demand.Based on a large number of historical data sets accumulated in Ningbo ride-hailing market,this study accurately predicted the short-term supply and demand state of ride-hailing,providing basic support for ride-hailing supply-demand matching and dynamic pricing.In addition,dynamic pricing is used to reconstruct the spatial and temporal distribution of demand and supply on the road network,which is conducive to reducing the empty driving distance,shortening the waiting time of passengers,improving the operating efficiency,and improving the travel comfort and accessibility of passengers.Firstly,based on the ride-hailing order data and GPS data,the passenger pick-up and drop-off location information was extracted,and the Dynamic Time Warping(DTW)algorithm was introduced to construct the spatio-temporal characteristics analysis model of ride-hailing travel demand based on density peak clustering to identify the spatio-temporal patterns of ride-hailing travel.This paper analyzed the relationship between ride-hailing travel patterns and urban functional areas from two aspects of travel intensity and travel spatio-temporal characteristics.The results show that the density peak method is significantly better than K-means and DBSCAN algorithms in terms of time series and spatial distribution clustering results.The travel intensity showed a hierarchical distribution,and the travel demand decreased from the central area of Ningbo Sanjiangkou to the outside,and the spatial structure of travel intensity was the central commercial and business district,the mixed commercial and residential district,and the residential district.Secondly,by comparing the advantages and disadvantages of a single prediction algorithm,a Stacking ensemble learning algorithm was introduced,and an e-hailing demand based on random forest,LightGBM and LSTM,and SVR as the second layer learner was proposed.The prediction results were analyzed for the statistical periods of 10 min,15min and 30 min,respectively.The results show that the Stacking integrated learning prediction method has high accuracy and good applicability,and the effect is most significant when the partition time is 30 min.Thirdly,combined with the data of supply and demand ratio of ride-hailing vehicles,the model was constructed layer by layer and the parameters are refined by deep learning,and the short-term supply and demand ratio prediction model of online hailing vehicles were built based on the characteristics of time difference and spatial dependence.In the case of considering the spatio-temporal characteristics,the spatio-temporal three-dimensional data was taken as the input of CNN,and the input data is defined according to the spatial correlation.The strong memory ability of LSTM is helpful to improve the accuracy and stability of the prediction model.The results show that the short-term supply and demand ratio prediction model of ride-hailing based on CNN-LSTM is reasonable.The prediction effect is better than the single time series prediction model,which can be effectively applied to the short-term prediction of the actual supply and demand ratio of online hailing vehicles.Fourthly,based on the demand prediction of online car-hailing,the data characteristics of online car-hailing waiting time and questionnaire survey,this thesis established an online car-hailing dynamic pricing method driven by waiting time data,combs the key influencing factors of online car-hailing pricing,constructed a decision tree model,and intuitively divided the waiting time of online car-hailing passengers into 11 categories.Based on the static pricing of online car-hailing,the influencing factors of online car-hailing price were added to determine the dynamic pricing mode of online car-hailing for each category,and a comparative analysis was made before and after the dynamic pricing.The results show that when passengers are less tolerant of waiting time,the online car-hailing price can be reduced to reduce the cancellation rate of orders.When passengers have a high tolerance for waiting time,the online car-hailing price can be increased to increase the income of online car-hailing platform and drivers.Before and after dynamic pricing,only a small proportion will increase,and the total revenue and total profit increased by 4.1%,while the proportion of orders with waiting time less than patient waiting time increased by 25.8%.Finally,the interactions between dynamic pricing,platform operation strategy and passenger travel behavior were sorted out,dynamic pricing,waiting time and service level are introduced into the macro-operating service system of online taxi,the logical relationships among key elements were analyzed,the systematic influence mechanism of online taxi platform service strategy driven by dynamic pricing was identified,and a system dynamics-based dynamic pricing influence effect assessment model is constructed.The system dynamics model was validated by combining the prediction results of the demand-supply differential in Ningbo and other key variables with the measured data,and its validity was verified.The results show that the dynamic pricing can "cut the peaks and valleys" and further make the supply and demand of online taxi reach the equilibrium state,when the price of online taxi fluctuates with the supply and demand,it can meet the overall profit increase of online taxi and make the supply and demand of online taxi reach the equilibrium state,which has a positive effect on the operation service strategy of online taxi platform.This has a positive effect on the operating strategy of the online taxi platform.