Risk Identification Of Motorway Sections Based On Multi-source Data

With the acceleration of urban development process,people’s travel demand rises,the contradiction between supply and demand of roads gradually emerges,the highway shows the normalization of traffic congestion,the phenomenon of frequent traffic accidents,the road traffic safety situation is serious,the urgent need for research on active prevention and control of traffic risks.Therefore,based on multi-source data,this paper analyzes the correlation between the factors influencing accidents on the highway accident black spot section,explores the causes of traffic operation risks,explores the traffic operation risk characteristics,and realizes the discrimination of traffic operation risks on the highway section.The research can help relevant departments to grasp the road traffic operation risk situation in real time and reduce the accident rate.Firstly,the paper takes the 30 km road section of Kunshi Expressway as the research object,divides the road section unit based on the historical accident data and the moving step method,identifies the accident black spots by combining the Relim algorithm and the equivalent accident method,and explores the correlation between the influencing factors of accidents by using the Apriori algorithm to investigate the causes of accidents.It is found that the formation of accident black spots is mainly influenced by the road environment and drivers’ lane changing and following behaviors,which helps to analyze the dynamic risk of road sections and effectively explore the risk characteristics of traffic operation.Secondly,the acquisition of multi-source data of highway is introduced,and the traffic information of accident black spot section is collected by multi-detector,and video traffic information is extracted by multi-scale KCF algorithm and YOLOv7algorithm;the acquired data are classified and identified and processed abnormally,and the same type of data are fused based on Kalman filtering method.After Bayesian risk detection,the Bayesian risk value of the fused data is reduced by 15% on average compared with the original data,indicating a better effect of data fusion,which provides a more complete and accurate data set for traffic operation characteristics analysis and risk discrimination.Then,by analyzing the macro traffic flow characteristics and micro risky driving behavior,the risk characteristics of traffic operation of the road section are explored,and based on the three-phase traffic flow theory,the mechanism of risk generation in different traffic states is explored.It is found that the traffic flow of the accident black spot section has large speed dispersion and mismatch between headway and speed;and the driver’s following and lane changing behavior is an important factor that leads to the turbulence of the traffic flow state.This study provides a certain basis for effectively grasping the law of traffic operation risk generation and accurately identifying risk characteristics.Finally,the paper improves the fuzzy c-mean clustering algorithm based on the independence weight coefficient method,and selects speed,headway spacing,flow rate,lane change ratio and large vehicle ratio as the risk feature parameters for traffic risk state clustering,and determines their risk levels based on the assigned weight values of the feature parameters.Then,the traffic risk state discrimination model based on the gradient boosting tree is constructed and compared with the random forest and K-neighborhood models to verify the effectiveness of the proposed model.From a comprehensive view,the accuracy of the gradient boosting tree model is slightly higher than that of the above-mentioned models in each state,which confirms the effectiveness of this method for the risk discrimination of highway sections.Then,the accuracy of risk level discrimination in practical application is improved based on risk superposition,which provides some theoretical basis and technical support for active prevention and control of traffic risks.