Research On Regional Traffic Flow Prediction Based On Floating Car Data

With the advancement of urbanization,the number of urban motor vehicles has increased year by year,and the contradiction between transport capacity of urban road network and traffic travel demand is increasingly prominent,traffic congestion has become one of the most serious problems in urban transportation,it has brought great inconvenience to the travel of citizens,and also caused serious adverse effects on the urban environment and economic development.Compared with traffic congestion mitigation measures such as urban road expansion and reconstruction,vehicle restrictions,etc.,traffic flow prediction has the characteristics of low cost and easy implementation.Through accurate prediction of traffic flow in urban areas,citizens can effectively avoid congested road sections,and traffic management departments can also formulate more scientific and effective traffic control measures to improve the management ability of urban traffic and alleviate road traffic pressure.Traffic flow has complex spatio-temporal nonlinear characteristics.Due to its powerful data characterization capabilities and feature extraction capabilities,deep learning-based modeling methods have become the mainstream solutions in the current traffic flow prediction field.At the same time,with the development of sensors and internet of vehicles technology,floating car data collected by on-board GPS devices gradually enter the vision of researchers in the field of traffic engineering.Floating car data has the characteristics of rich data volume and wide spatial distribution of data,the research on regional traffic flow prediction based on floating car data is low-cost and more flexible.Therefore,constructing a suitable regional traffic flow prediction model based on the deep learning modeling method of floating car data is one of the focuses of current traffic flow prediction research.In view of this,this thesis takes floating car data as the research object,proposes a floating car data preprocessing scheme;comprehensively considering the multi-period spatio-temporal characteristics,image-like characteristics of traffic flow grid data,and regional POI spatial facility information,combined with deep learning,a regional traffic flow prediction method system is proposed,details as follows:(1)Aiming at the floating car dataset used in this thesis,a corresponding data preprocessing scheme is proposed.Using the grid division method to divide the research area into a grid;using the ray map matching method combined with the enveloping rectangle to match the data record of the floating car to the corresponding position grid,and obtain the grid number of the position where the data record of the floating car is located;Finally,based on the grid number and the time slice sequence number obtained by time slice division,the traffic flow grid data covering the entire urban area is calculated.This data preprocessing scheme lays the foundation for the subsequent analysis and forecasting of regional traffic flow.(2)Aiming at the spatio-temporal periodic characteristics of traffic flow grid data,combined with deep learning,a regional traffic flow prediction model based on multi-period spatio-temporal components is constructed to complete the accurate prediction of urban regional traffic flow.Among them,the multi-period spatio-temporal component is composed of Conv LSTM and MRes Net,Conv LSTM ensures the consistency of spatio-temporal feature extraction of regional traffic flow,and MRes Net completes the refined feature re-extraction of regional traffic flow.(3)Aiming at the unique image-like characteristics of traffic flow grid data,and the data dynamic relationship between regional POI spatial facility information and traffic flow,a regional traffic flow prediction model considering grid image-like characteristics and regional POI is proposed.The model use 3D CNN to construct a multi-period spatio-temporal feature extraction module of regional traffic flow,fully capture the multi-period spatio-temporal correlation of traffic flow;use QGIS professional geographic information software to obtain and process six major regional POI grid data in the research area,build a regional POI feature extraction module based on 2D CNN to capture the dynamic data correlation between regional POI spatial information and traffic flow.This research provides a systematic data preprocessing scheme and model modeling ideas for the regional traffic flow prediction based on floating car data,which promotes the development of research on traffic flow prediction in large areas such as cities,and provides a theoretical basis and decision-making basis for the governance of urban traffic congestion.