Analysis Of Influencing Factors And Demand Prediction Of Online Car-hailing Travel In Hotspot Areas Based On Built Environment

As an important part of urban transportation,online car-hailing plays an important role in the travel of residents,and to a certain extent,it has alleviated the problem of "difficult to get a taxi",but the problem of uneven supply and demand still exists.Based on the order data,POI data and weather data of Xi’an city,this paper explores the passenger-carrying hotspots during the peak period,analyzes the travel influencing factors,and constructs a prediction model for the travel demand of online car-hailing,which can not only help online car-hailing platforms to dynamically adjust vehicle supply,optimize operation and management decisions,but also improve users’ travel experience and reduce waiting time.Firtsly,this paper cleaned and filtered the original data,visualize the spatial and temporal characteristics of online car-hailing travel using python and Arc GIS,and analyze their distribution characteristics.After that,the improved DBSCAN spatial clustering algorithm based on k-distance is used to find the peak periods of weekdays and rest days,and to refine the location of the hotspots.Secondly,the factors influencing online car-hailing travel are analyzed.To make the study reliable,the concentrated distribution range of passenger-carrying hotspots is taken as the study area.With travel demand as the dependent variable and multiple POI densities reflecting the built-up urban environment as the independent variables,the multicollinearity among the independent variables is tested,and stepwise regression is realized using SPSS software to screen out the explanatory variables for each peak hour and analyze their spatial autocorrelation,and all variables are found to be spatially positively correlated,meeting the model construction conditions.The least squares regression model and the geographically weighted regression model are constructed to compare the goodness of fit of the two,and the feasibility of the geographically weighted regression model for the study of travel influence factors is verified,and the spatial and temporal heterogeneity of online car-hailing travel influence factors is analyzed by taking the peak hours of weekdays and rest days as examples.Finally,based on the Tensor Flow2.0 deep learning framework,Ji Xiang Village is selected as a typical passenger-carrying hotspot area,and a Bi-LSTM-based online car-hailing travel demand prediction model is constructed for weekdays and rest days.Combining the POI influencing factors with time factors and weather factors as variables input into the model,the model is parameter optimized and the model structure is determined.RMSE,MAE and MAPE are selected as model evaluation indexes,and the Bi-LSTM model is compared with several traditional models which shows that the Bi-LSTM model has the highest prediction accuracy and the prediction model performance is more stable under different date attributes.The validity and smoothness of the Bi-LSTM prediction performance are verified by analyzing the model errors for different peak hours.