GPS Data-driven Traffic State Analysis And Congestion Pattern Forecasting Of Urban Road Network

Taxi trajectory data in the city can reveal the changes related to the traffic operation state of the road network and travel behavior.Data mining methods such as statistical theory,machine learning,and deep learning have promoted in-depth research on taxi GPS data in the transportation field.And further promote the research of urban road network traffic operation state analysis,operation state evaluation and prediction theory,so as to provide reference for travelers to choose travel modes and routes.The traffic management department will have a better understanding of the changes in traffic conditions,so that it can master the change rules of traffic conditions and implement effective traffic management to alleviate traffic congestion.The paper takes "pre-processing of taxi GPS data and road network,construction of comprehensive evaluation index of traffic running state,analysis of traffic running state,construction of traffic congestion mode prediction model,and case study" as the main line to carry out research on traffic running state analysis and prediction theory.The main research contents of this paper are as follows:(1)Based on taxi GPS data,the evaluation system under different road grades is constructed,and the multi-index fusion algorithm based on direct fuzzy theory is proposed for the calculation of comprehensive evaluation indexes.Firstly,the selection principles and standards of evaluation indexes are determined from electronic tag/license plate recognition,GPS floating vehicle data,traffic management bureau and road traffic sensors.Secondly,by analyzing taxi GPS data and fuzzy C-means clustering algorithm,a comprehensive data model combining taxi GPS track and traffic state is built to provide data support for subsequent research.Accordingly,traffic running states of different levels of roads are constructed based on the number of taxi tracks and the average speed of each track.Then,based on the basic evaluation indexes,the direct fuzzy theory is used to construct the evaluation system under different road grades,and the comprehensive evaluation indexes of highway,arterial road and secondary trunk road/branch road are determined.(2)Based on the evaluation system under different road grades,the evaluation algorithm of traffic congestion mode is constructed,and the multi-level urban traffic operation state is analyzed from the perspective of time and space.Firstly,the difference of spatial-temporal relationship between sections is analyzed by using data mining method.Then the Euclidean distance and statistics theory are used to analyze the traffic status at different levels from spatial-temporal respectively.Finally,the traffic congestion mode and traffic running state are analyzed from three aspects of micro,meso and macro,so as to verify the feasibility of the comprehensive evaluation index to grade and evaluate the traffic running state.(3)Based on the variation characteristics of spatial and temporal correlation of the observed values of the comprehensive evaluation index,the LSTM neural network prediction model embedded with spatial and temporal correlation variables was built.Firstly,the prediction model based on LSTM is determined.Then,the input variables of the LSTM model are constructed from the perspectives of time,space and spatial-temporal,and the corresponding threshold criteria are given.In order to analyze the influence of time,space and space-time relationship on the prediction accuracy,T-LSTM model,S-LSTM model and TS-LSTM model were used to carry out the comprehensive evaluation index prediction research on different road sections.Finally,the prediction results of the TS-LSTM model are compared with the traditional machine learning model SVM and the basic deep learning model LSTM.The prediction results show that the TS-LSTM model fully combines the advantages of statistical methods and deep learning,has good prediction performance,and is suitable for the prediction of traffic congestion patterns.