Analysis Of Recurrent Congestion Propagation Patterns And Congestion Prediction Based On Taxi GPS Data

The continuous growth of motor vehicle usage and the increasingly serious separation of residents from work and residence in China have led to the worsening of urban traffic congestion,which hinders the development of cities and the improvement of people’s living standards,therefore,the research on traffic congestion is a topic of continuous and in-depth attention by domestic and foreign scholars.The current research focuses on congestion identification,congestion propagation and congestion prediction,among which the research on congestion propagation only focuses on the identification of propagation paths and fails to quantify the propagation possibility.This problem is challenging for the precise management and accurate prediction of congested areas,so it is necessary to introduce advanced models to study the congestion propagation probability and optimize and iterate the congestion prediction methods.This paper aims at solving the urban traffic congestion problem.And the paper applies the taxi GPS data to the main line of "spatio-temporal cube model construction-recurrent congestion area identification-congestion propagation pattern analysis-congestion prediction",and carries out the research on urban traffic congestion.The study provides a reasonable basis for congestion management by traffic-related departments.Firstly,the spatio-temporal cube model is constructed,and the spatio-temporal scale of the model is determined by the proportion of invalid grid and the average number of trajectories of the effective grid,and the position parameters of the grid are adjusted to finally realize the spatio-temporal mapping of taxi GPS data.Secondly,the number of vehicle trajectories and the weighted average travel speed index are selected to jointly identify traffic congestion areas,and based on the relative spatio-temporal stability of recurrent congestion,a time-phased recurrent congestion grid identification method is proposed to refine the identification of the specific time period when the grid is formed by recurrent congestion.Again,using the STC algorithm,a congestion spatio-temporal propagation tree is established.Based on this,a frequency-weighted frequent propagation relationship set mining method is proposed to reconstruct the frequent congestion propagation subtree for the dynamic nature of traffic congestion propagation,followed by the introduction of a dynamic Bayesian grid to establish a congestion propagation model and learning to obtain the congestion propagation probability and quantify the congestion propagation possibility among grids.Finally,traffic congestion prediction models are constructed based on single deep learning models such as LSTM model,GRU model and fused deep learning models such as LSTM-GRU model,GRU-LSTM model and LSTM2L-GRU model,respectively.The models take into account the temporal,external and internal factors affecting the congestion state of the grid in the selection of input variables.By comparing the prediction accuracy and running time of each deep learning model,it is concluded that the GRU-LSTM model is the best traffic congestion prediction model with 90% prediction accuracy and short running time.This study can provide traffic-related departments with early warning information of congestion,and provide some theoretical basis for them to formulate congestion relief measures and make early intervention,which is of practical significance.