Online Car-hailing Travel Demand Forecasting In Pick-up Hotspots Areas Based On Order Data

With the rapid development of the Internet economy,online car-hailing has become one of the important modes of transportation for people’s daily travel.With its "door-todoor" service,online car-hailing not only meets the requirements of residents for high travel quality,but also solves the travel needs of residents in areas not covered by public transportation.However,due to the randomness of residents’ travel,there are some problems in online car-hailing operation,such as high empty driving rate of drivers and passengers "unable to get on the bus".Based on online car-hailing order data,this paper analyzes the characteristics of online car-hailing travel demand by applying the improved DBSCAN algorithm to mine the travel hotspots to establishes the travel demand prediction model of single area and multi area.Finally it selects the travel demand data of hot spot to analyze the prediction effect of the model,which verifies the effectiveness of the model.The research in this paper is of great significance to improve the operation and dispatch level of online car-hailing platforms and alleviate the contradiction between supply and demand.This paper mainly completes the following work:(1)Characteristic analysis of online car-hailing travel demand and identification of related influencing factors.The paper clarified the definition of the research related issues in the paper,and preprocessed the data;analyzed the time characteristics of travel demand from different week attributes and different time periods.By comparing the spatial distribution of travel demand and pick-up area in different working day attributes,different periods,the spatial distribution characteristics of travel demand in different periods are analyzed.By comparing the regional network car-hailing inflow and travel demand,the time-dependent characteristics of travel demand and historical inflow are analyzed.The correlation coefficient is used to analyze the factors that may affect the travel demand for online car-hailing,and the input characteristic variables of the prediction model are determined.(2)Mining hotspots of online car-hailing travel.Based on the analysis of the spatialtemporal distribution characteristics of online car-hailing travel demand,the parameters of DBSCAN algorithm are optimized,and the travel hotspots in different peak hours of working days and non-working days in the study area are mined,and the hotspot index is introduced to divide the hotspots into continuous,frequent and occasional travel hotspots.(3)The travel demand prediction model of single area and multi area are established.Aiming at the forecasting the travel demand in a single area,considering the time characteristics of online car-hailing travel demand,a genetic algorithm optimized Attention Long Short-Term Memory(GA-ALSTM)travel demand forecasting model was established.Aiming at the forecasting the travel demand in multi area,considering the spatio-temporal characteristics of online car-hailing travel demand and the time dependence of historical inflow,an Attention Convolutional Long Short Term Memory(AConv LSTM)travel demand forecasting model is established,which can be applied to the grid division mode.In view of the defect that convolutional neural network can only deal with orderly grid structure,an Attention Graph Convolutional Long Short Term Memory(AGC-LSTM)travel demand forecasting model is established,which can be applied to irregular area division.Finally,the effectiveness and applicability of the model method proposed in this paper have been verified through analyzing the travel demand data of the continuous travel hotspot area with different evaluation indicators.There are 62 figures,31 tables,and 84 references in the paper.