Research Of Traffic Conflict Models Between Non-motor And Motor In The Intersection Based On Extreme Theory

The improvement of urban traffic safety is an important direction for sustainable development and environment improvment.Among them,non-motor vehicle users,as a traffic vulnerable group,is easily to be injured in traffic accidents.However,researches mainly focuse on motor vehicles and pedestrians,while is less concerned on non-motor vehicle users.On the other hand,traffic conflict technique inevitably has the problem of the correlation between traffic conflict-traffic accidents,while ignores the objectivity of traffic conflict judgment.However,traffic accident data is too small and the accident environment always has uncontrollable changes.Based on the extreme value theory,this paper conducts traffic conflict models between nonmotor and motor in the intersection based on extreme theory.Firstly,a literature review was conducted in three main respects: non-motor vehicles-motor vehicle accidents,traffic conflict and extreme theory.Then,accidents classification methods,research methods on traffic conflict and the application of extreme theory on accident analysis were summarized,so as to build a theoretical base for this paper.Secondly,this paper proposed a In-vehicle method for identification of traffic conflicts between non-motor vehicle and motor vehicle.Based on three types of traffic conflicts: potential conflict,scraping conflict and traffic accident,specific PET identification was proposed in order to improve the existing identification method.Then,based on the generalized extreme value distribution and generalized Pareto distribution assumptions,two kinds of non-motor-vehicle conflict prediction models,namely the block maxima method(BMM)conflict model and the peaks over threshold(POT)conflict model,are constructed to achieve the quantitative prediction of the conflict extremum at each intersection.Targeted adjustment methods are proposed for two key problems of conflict prediction in introducing extreme value theory :(1)nonindependent problems that may occur in urban intersection conflict data.For the BMM conflict model and POT conflict model respectively,interval adjustment and traffic string division method are formulated;(2)the quantitative selection of the over-threshold conflict model.A combination of qualitative graphic method and quantitative parameter method is adopted,so as to quantitatively select thresholds.Using the video record conflict data of 20 intersections in Nanjing city as examples,the BMM extremum model and the POT extremum model are constructed respectively,with the model fitting tests which ensured the feasibility of the prediction model.Based on the extremum theory,the traffic conflict extremum of different intersections in different recurrence periods is predicted,and the safety level of different intersections is quantitatively evaluated and compared.In order to further improve the applicability of the prediction conflict models,based on the BMM models,this paper constructs a comprehensive traffic conflict model by using covariates,so as to realize the comprehensive prediction of multiple intersection conflicts.Alternative covariate come from the intersection characteristic variables in environmental,traffic structure and land use dimension.Based on the location,shape and scale parameter in BMM conflict models,the stepwise regression method,Lasso regression and random forest-Boruta algorithm are used to identify the corresponding variable.Based on the above three methods,the corresponding covariate model is constructed,and the covariate modeling method is evalued by RMSE,MAE and AIC.Based on the optimal covariant modeling method,the corresponding comprehensive prediction model is constructed.Then,the BMM confilct model,the POT conflict model and the comprehensive model are compared from three aspects: data utilization efficiency,prediction reliability and prediction accuracy.The results show that the prediction effect of the BMM confilct model is optimal,but the prediction reliability is relatively low.Compared with the BMM confilct model,the comprehensive model has less loss of prediction ability and has certain prediction effect.Based on the quantitative extreme prediction,a spatial modeling method of urban intersection perspective in the micro scale are constructed.The random forest method is used to build the corresponding prediction model,so as to realize the prediction of intersection conflict’s location and shape.Then,based on the position accuracy and shape similarity,a quantitative comparison method is constructed.Intersection I2 are analyzed as an example,and the results show that the prediction model for(1)turn left into the crossing vehicles within the non-motor vehicles crossing the street(through traffic lights)conflict,(2)the motor vehicle road turn right turn right lead to motor vehicle two class conflict and conflict on the left side of the crossing go straight nonmotor vehicles position have better prediction effect;Compared with the actual area,the position error is relatively small.The shape similarity can be further optimized.It is of certain reference significance to predict the intersection conflict area by using the micro-vision-random forest method.