Research On Urban Traffic Congestion Identification And Prediction Based On Taxi GPS Data

Traffic congestion is a common traffic problem faced by major cities in China.Generally,due to the different road widths,signal control and road driving rules of each city,each city will face traffic congestion problems,and the time period will be different.The control and complete governance of this problem have not been completely resolved,and it still affects the city’s transportation.The more traditional congestion problems are usually regular,for example,the frequent occurrence time period is during the morning and evening peaks of travel.Therefore,through the occurrence of congestion in some road sections,the road traffic congestion status can be predicted with the help of roads and time and space features,which is convenient for the government.The management department quickly responded to alleviate urban congestion.This research is based on the main line of research on "processing,analysis,mining and utilization of massive data",and makes full use of the trajectory data generated by the Ningbo taxi operating GPS terminal.First of all,through the analysis and processing of GPS taxi trajectory data and urban road network data in Ningbo,this paper models the research data to prepare for the subsequent congestion identification and prediction model input data.Based on the data that has been modeled,the time and space step parameters are determined and extracted based on the space-time data division model.Model the data on the basis of data preprocessing,and construct a traffic state recognition model inside the grid,and design a clustering algorithm to identify frequent congestion grids,which will lay the data and theoretical basis for the subsequent prediction of the traffic state in the grid.When the phenomenon of congestion within the grid is studied,the rapid increase and decrease of the traffic flow and the running speed of the vehicle in the grid will occur.Therefore,this paper selects the traffic flow and average speed indicators as grid congestion determination indicators to construct a congestion recognition model.Finally,identify the congested area through the constructed congestion recognition model.Based on this data,a combined logistic regression and gradient boosting decision tree combined prediction model LR2 GBDT is constructed to predict the short-term traffic congestion state inside the grid.Logistic regression(LR)is used to train the original features,such as week attributes,time series attributes,the number of vehicles in the grid,and the average driving speed.Then,the Gradient Boosted Decision Trees(GBDT)model was constructed.The input variable is the weighted feature result trained by LR,and the prediction result is output;finally,the existing prediction methods are compared and analyzed with the prediction model constructed in this paper.