Freeway Traffic Crash Risk Research And Judgement Considering Multi-dimensional Dynamic Features Interaction And Spatio-temporal Impact Propagation Analysis

In recent years,the number of motor vehicles in China has increased rapidly,and the construction of freeways has also entered a stage of rapid development.As a result,in addition to the convenience of life,there are security risks.The traditional freeway traffic safety management mainly focuses on "post-event" control,which is passive.Moreover,the traditional dynamic security prediction does not consider the interaction of feature dimensions enough.With the development of freeway to informatization and intelligence,the importance of active prevention and control is becoming more and more prominent.In order to provide some technical and theoretical support for the formulation and implementation of accident active prevention and control strategy of freeway safety management department,more accurately predict the occurrence of freeway traffic accidents and reduce the delay time caused by accidents,this paper takes our country freeway crash data as the research object,considers the multi-dimensional dynamic mutual interactions among traffic flow,weather,road,and time features as the breakthrough point,constructs freeway crash risk prediction model.And based on the real-time traffic flow data,the temporal and spatial impacts of crashes are analyzed,so as to effectively predict freeway crashes and reduce the possibility of secondary crashes.The main research work of this paper includes the following aspects:(1)Preprocessing of data related to freeway crashes and selection of alternative characteristic parametersBased on the crash data,the matched traffic sensor data,weather data,and road data collected from the Beijing section of the Beijing-Harbin Freeway,this paper extracted the non-crash data samples by using random sampling method,selected the appropriate feature variables used in the freeway crash risk analysis,including 15 traffic flow variables,5 road characteristic variables,3 weather characteristic variables and 2 time characteristic variables,which established the data basis for real-time accident risk influencing factor analysis and modeling.(2)Analysis of influencing factors of freeway crash risk based on XGboost-SHAP methodA real-time crash risk prediction model based on XGboost(e Xtreme Gradient Boosting)was developed.The model was interpreted and visualized using the SHAP(Shapley Additive ex Planations)interpreter to obtain the ranking of the importance of each feature on the impact of crash risk and to analyze the relationship between each feature and crash.The relationship among the features and the crash risk and the twodimensional interaction effects among the features were analyzed.The results show that the traffic flow variables,road,weather and time features all affect the crash risk,and there are interactive effects among the features,which lays a foundation for the real-time freeway crash risk modeling,which is the core of this paper.(3)Modeling towards freeway real-time traffic crash prediction considering multidimensional dynamic features interactionTaking the multi-dimensional dynamic mutual interactions among traffic flow,weather,road,and time features as the breakthrough point,and verifying whether weather,road features improved the accuracy of real-time crash risk prediction model,this paper constructed multiple datasets and applies the Deep & Cross Network(DCN)algorithm to real-time crash risk prediction research for the first time,and compared it with logistic regression,support vector machine and random forest.The results show that considering the interaction of multi-dimensional dynamic features can effectively improve the classification performance of the model,and the area under c curve(AUC)of the DCN algorithm can reach 0.8562 on the dataset including traffic flow,time,weather and road features.(4)Analysis of temporal and spatial impact propagation of freeway crashesBased on the collected traffic flow data,this paper analyzed the speed changes before and after the crash,selected the speed change rate as the measurement index of the impact of the accident,constructed the speed change rate contour diagram using bilinear interpolation method,and fitted the outer contour of the spatio-temporal area affected by the crash according to the Savitzky-Golay filter fitting method.Finally,the quantitative indexes of the spatio-temporal impact of the crash were calculated and analyzed when the threshold value of the speed change rate was 20%,30% and 40% respectively,including the beginning time of the crash impact,the end time of the crash impact,the duration of the crash impact,the nearest distance of the crash impact,the furthest distance of the crash impact,the space of the crash impact,the spread speed of the crash impact and the dissipation speed of the crash impact.The results show that the smaller the speed change rate threshold is,the earlier the crash effect starts,the later the crash effect ends,the longer the duration of the crash effect,and the longer the distance of the crash effect.In addition,under different speed change rate threshold,the variation trend of the crash effect propagates with time is different.The spatio-temporal impact propagation analysis method used in this paper can quantify the spatio-temporal impact of motorway accidents more effectively.