Accident Analysis And Prediction Method Of Expressway Based On Whole Factors Of Driver,Vehicle,Road And Environment

Accident analysis and prediction,as an important task to ensure the traffic safety of expressway,the reasonableness of its method is not only related to the accuracy of the safety audit results of expressway,but also directly affects the efficiency of dealing with the traffic safety problem of expressway.As a complicated dynamic operation system,the traffic accident of the expressway is closely related to factors,such as driver,vehicle,road and environment,and the interaction influence between these factors.Restricting by the extraction of dynamic feature of factors,some factors,such as environmental factors,road spatial geometric features and so on,have not been accurately quantified,which makes the characteristic parameters used for analysis have obvious discreteness.As a result,the correlation accident analyses stay at the level of qualitative analyses,which reduces the reliability of their results.At the same time,tranditional accident analysis and prediction ignore the essence of the integrated dynamic coupling system of driver,vehicle,road and environment,which makes the results of accident prediction and analysis more one-sided.Some defects caused by factors' coupling effect may difficult to be found,which resulting in omissions in the analysis of accident causes,and finally reducing the efficiency of the management of traffic safety problem.In order to make up for the shortcomings of the traditional accident analysis and prediction methods in the dynamic feature extraction of factors and the integrated analysis of each factor,and to improve the accuracy and reliability of the accident analysis and prediction results of the expressway,this paper establishes an new accident analysis and prediction method based on whole factors from the essence of the driver-vehicle-road-environment dynamic coupling system of the expressway traffic system.Firstly,according to the biological characteristics of driving behavior pattern,a theoretical model of reflective arc accident analysis based on information transmission is proposed.Secondly,through theoretical research and field driving test,the real-time dynamic characteristic parameters of each factor in the coupling system are extracted by means of three-dimensional spatial differential geometry calculation,image recognition,psychological index collection and vehicle fault detection,and the whole factor evaluation index set for expressway accident analysis and prediction is established.Thirdly,the expressway accident prediction model based on Poisson regression model and negative binomial regression model is constructed by using the whole factor evaluation index set and the statistical distribution characteristics of expressway accidents.The reliability of the prediction model is verified by empirical analysis.Then,according to the parameter estimation results of these accident prediction models,the characteristic index set which has a significant impact on the accident is obtained.According to the theoretical model of reflection-like arc accident analysis,the characteristic index set which significantly affects the accident is divided into three levels: stimulus information,perceptual response and behavior appearance,and a VAR model is constructed to analyze the interlayer relationship of the characteristic index.The method of correlation analysis of influencing factors of expressway accidents under the condition of whole factors is formed.Finally,the Granger causality test method is used to discuss whether the layerby-layer information transmission structure of expressway section is stable.Based on this test,combined with the pulse function analysis and error variance decomposition analysis in VAR model,the causes of accident-prone sections are deeply analyzed,and the cause analysis method of accident-prone section based on Granger test is put forward.The comprehensive innovation of accident analysis and prediction method in theoretical domain,data domain,method domain and analysis domain is realized.The main research contents and innovative results are as follows:(1)The accident of expressway is affected by the driver,vehicle,road and environment.It is the key to solve the traffic safety problem of the operation highway to carry out the accident impact analysis based on whole factors,and the premise is to extract the characteristic index of whole factors effectively.In this study,the basic data are collected by naturalistic driving test.Specifically,the driving video is processed by image recognition technology,from which the characteristic indexes such as sky proportion and luminance are obtained to characterize the environmental factors.The accurate calculation method of three-dimensional spatial differential geometry is established to obtain the spatial curvature and spatial torsion which could present the essence of the spatial curve of road alignment.These two spatial geometry indices,together with the slope index,are used as the characterization parameters of road factors.The vehicle factor is characterized by indices,such as speed,acceleration,fuel consumption and other real-time driving state information,which are obtained by the vehicle's on-board diagnostics,and the vehicle lane departure index which is obtained from driving video by image recognition technology.Driver's fixation,heart rate and skin conductive response are collected to represent the driver factor.On the basis of successfully extracting all kinds of characteristic indices of whole factors,a fixed length method is used to divide the road section unit,then,the statistics of 18 characteristic indexes in the road section unit are determined as the evaluation indices.Thus,an evaluation indices set of whole factors is established,which treat the road section unit with fixed length as the analysis object.(2)Based on the evaluation indices set of whole factors,the statistical characteristics of the accident are investigated when regarding the occurrence of the accident as an independent event.Poisson regression model and negative binomial regression model are established as the accident prediction models,in these two models,the evaluation indices of whole factors are regarded as the independent variable and the number of accident is regarded as the dependent variable.The empirical analysis results of these two models in the testing road is obtained by using Stata 15.0,the results show that Poisson regression model and negative binomial regression model have good accuracy and stability under the evaluation indices set of whole factors.On the whole,the prediction result of negative binomial regression model is better than that of Poisson regression model.However,the evaluation indices which have a significant impact on the accident according to the parameter estimation results of Poisson regression model are wider than those of the negative binomial model.Moreover,the parameter estimation results of Poisson regression model and negative binomial regression model show that mean duration of fixation,proportion of fixation in road-ahead area of road unit,mean heart rate,max lane departure,mean fuel consumption,maximum difference of sky proportion,maximum difference of luminance,mean_?,?? and mean longitudinal acceleration all have significant effects on the number of accident,and the driver factor has the greatest influence on the accident.(3)According to the evaluation indices which significantly affect the number of accident,the corresponding characteristic indices with distance sequence characteristics are selected.In order to analyze the correlation between the characteristic indices,the characteristic indices are divided into stimulus information layer,perceptual response layer and behavior appearence layer,in which the characteristic indices of stimulus information layer are spatial curvature,spatial torsion,sky proportion and luminance;the characteristic indices of perceptual response layer are heart rate,skin conductivity response,proportion of fixation in road-ahead area and fixation duration;and the characteristic indices of behavior appearence layer are velocity,longitual acceleration,fuel consumption and lane departure.Then,a layer-by-layer vector autoregression model(VAR model)is established to analyze the correlation between the indexes of each layer.According to the statistical analysis of the accident of the test road,the test road is divided into two categories: the normal section and the accident-prone section.An empirical analysis of VAR model is carried out based on the normal section of the accident in the test road by using Eviews 10.0.The results show that the stimulus information layer has the ability to predict the characteristic indices of the perceptual response layer,and the perceptual response layer has the ability to predict the characteristic indices of the behavior appearence layer.The layer-by-layer VAR model is stable,and the impulse response function and variance analysis of VAR model show that the contribution of the characteristic indices in upper layer to changes of different characteristic indices in lower layer is different.(4)Combined with the biological characteristic of driving behavior mode,a layer-by-layer information transmission model of quasi-reflection arc is proposed,and Granger test method is used to judge whether the layer-by-layer information transmission is successful or not,which is used as a starting point to analyze the causes of accident-prone section.Thus,an analysis method of accident-prone section under the condition of whole factors is formed based on Granger test.An empirical analysis based on the test road is carried out.Firstly,according to the results of Granger test in the normal road section,the normal type of layer-by-layer information transmission structure is determined.Secondly,the layer-by-layer information transmission structure of the tunnel accident-prone section and the curve accident-prone section in the test road is determined by Granger causality test.Finally,the layer-by-layer information transmission structure of the accident-prone section and that of the normal section is compared and analyzed,and combined with the results of impulse response function analysis and variance analysis in the VAR model,the causes of the formation of the accident-prone section are obtained.The main research results of this paper provide a new theoretical and methodological support for improving the traditional accident prediction and analysis methods of expressway and improving the traffic safety situation of expressway,and lay a solid theoretical foundation for the comprehensive management of accident-prone section.