Research On Taxi Travel Demand Forecasting Model Based On Hot Areas

As a special public transportation mode,taxi is favored by more and more short-distance travelers due to its fast,convenient,comfortable and safe features.However,with the development of the taxi industry,the imbalance between supply and demand caused by the uneven distribution of urban vehicles and passenger flow is especially serious.In order to solve this problem,it is important to construct a forecasting model of taxi travel demand under the influence of multiple factors so that taxi operating companies can make scientific and reasonable scheduling arrangements for taxis in advance.This paper takes the GPS data of more than 10,000 taxis in Xi'an in April 2017 as the research object.The pseudo-distributed processing framework written in Python is used to complete the processing of large-scale GPS original data,including data cleaning,coordinate transformation,map matching,OD extraction,etc.,and the validity and convenience of the framework are proved by the comparison of different parameters.Secondly,a density field-based hot spot detection model is proposed for hot spot area mining.Compared with other clustering models,this model can adjust parameters according to different research scales,which greatly reduces the difficulty in parameter selection.By using this model and combining statistical methods to discuss the spatial and temporal distribution characteristics of taxi hotspots,the study shows that the travel peak time of taxi is slightly different from the travel peak time of residents,which is 6:00~8:00 in the morning,12:00~14:00in the afternoon,and 20:00~22:00 in the evening;The travel time distribution of each time period is relatively scattered,most of them are concentrated in 5 to 25 minutes,and most of them are short trips of 5 to 10 minutes;in the spatial distribution,the hotspots of getting on the taxi mainly distribute in the traffic service area and nearby the urban arterial roads;according to the urban functional areas,the morning peak hour hotspots are higher than afternoon and evening peak hourin residential areas;the hot spots in the medical service area have higher hot spots level in the morning and afternoon peak hours,and the hotspots in the same period of time tend to have higher level in workdays than in weekends;the hotspots level in large commercial service areas shows that afternoon and evening peaks have a higher level than morning peaks.However,due to the different locations,the hotspots in other small commercial service areas are quite different from each other.Finally,two different functional areas were selected as the experimental group,and the major urban area was taken as the control group to construct.The set of factors influencing taxi travel demand was constructed by combining weather,air pollutants and other factors.The ridge regression model,the random forest regression model and the combined prediction model based on the weight of the two models were applied respectively to predict the taxi demand in the experimental group and the control group.The research results show that the performance of the model is influenced by the study area and the evaluation index,from the perspective of the study area,the prediction effect of major urban area is better than other research areas,take R~2as an example,three kinds of modelsapplied in major urban area are all above 0.90,while the other two of study area's models at around 0.80;From the perspective of regression model,the best model selected by different evaluation indexes is not the same,but the random forest regression prediction model is rated as the best model most times.In some ways,the accuracy of the random forest regression prediction model is better than the other two models.