Deduction Algorithm And Case Study Of Expressway Traffic Situation Under The Influence Of Accident

The highway network is the lifeblood of economic development.In order to reduce the negative impact of congestion caused by traffic accidents,it is necessary to dynamically analyze and predict the evolution of the traffic situation after the accident and understand and visualize the evolution of the traffic situation accurately,in order to assist the management department of expressway to carry out situation research and assessment,and take effective management methods and congestion control.The influence of traffic accidents on traffic situation deduction is diverse and complex.The neural network methods based on traffic big data provide a new modeling idea for traffic situation research.In this thesis,by constructing a traffic flow prediction model that considers both the inherent characteristics of traffic accidents and spatial-temporal characteristics,a traffic situation deduction algorithm under the influence of accidents is proposed,so as to provide guidance for traffic guidance,information services and traffic management.Firstly,by sorting out the research status of related research,the research goal of this thesis is established: to build a traffic flow prediction model by integrating spatialtemporal features,to explore the impact of features of accidents on traffic situation,finally,obtain the construction idea of the traffic flow prediction model and situation deduction algorithm.At the same time,the technical route and main research contents of this thesis are given.Secondly,this thesis describes the traffic flow data and accident data of expressway,through missing data repair,data smoothing and normalization and other preprocessing methods to improve data quality.According to the spatial-temporal features of traffic speed,its inherent spatial-temporal correlation is analyzed.According to the accident situation,the influence of accident grade,accident location,weather and accident type is analyzed,so as to provide guidance for the follow-up introduction of accident features.Thirdly,aiming at the problem of traffic flow speed prediction,this thesis constructs a neural network model considering the fusion of spatial-temporal features based on Seq2 Seq architecture(DSTMA).The DSTMA model fully extracts the spatial-temporal features of multiple traffic nodes through spatio-temporal embedding module,spatiotemporal attention module and transform attention module.By using dense connection structure among each module,the feature reuse is realized and the problem of gradient disappearance during training is avoided.The experimental results on two kinds of expressway traffic flow datasets show that DSTMA achieves better prediction accuracy for the prediction of traffic flow speed within 1 hour.Then,considering the influence of the inherent features of the accident on the traffic situation after the accident,this thesis studies and constructs SAE network to mine the inherent influence of the accident on the traffic flow on the basis of the DSTMA model,and forms the Mix-DSTMA model to realize the multi-step prediction of the traffic flow speed after the accident.After the hidden features of accidents are obtained by SAE,a feature fusion module is constructed,and various fusion mechanisms are used to learn the abstract representation of the impact of accidents.The experimental results on the accident dataset show that the prediction result of traffic flow speed within 3 hours is more accurate than that of DSTMA,Mix-DSTMA after introducing accident characteristics.After obtaining the speed of each traffic node through the Mix-DSTMA model,combined with the spatial interpolation method and the division of congestion threshold,the traffic situation deduction algorithm under the influence of accident is established,thus the traffic situation representation and deduction of the whole road section is formed.The result of case analysis shows that the algorithm can deduce the traffic situation of the road section within 3 hours accurately and can describe the changing trend of congestion after the accident better.Finally,in order to realize the visualization of traffic situation deduction,this thesis designs a situation deduction system of expressway based on Web GIS system and frontend technical framework.The system can embed the relevant algorithms of traffic situation deduction,and through the click and interaction of users,the traffic network situation deduction process display and chart visualization of accident information can be realized at the front end.