Study On Traffic Conflict Prediction Model Of Freeway Interchange Merging Areas

The Interchange merging area serves as an entrance for the on-ramp vehicles to enter the freeway and is an important "valve" that assists the interchange to complete the vehicle turning function of the freeway network.However,at interchange merging areas,due to frequent vehicles’ merging behavior and the complex traffic environment,drivers are prone to misjudgment and conduct wrong operations,which may increase the probability of traffic conflicts,and even traffic accidents.Therefore,under the background of the increase in the number of interchanges and the outstanding safety issues at interchange merging areas,the study of the safety issues at interchange merging areas,the construction of safety analysis models,and the in-depth exploration of the safety influencing factors will be helpful and of great significance to improve the service quality of interchange areas.This paper intends to study the traffic conflict prediction model of freeway interchange merging areas and aims at revealing the mechanism of merging conflicts at the micro level.Firstly,the study analyzes the traffic operation characteristics of the Interchange merging areas and proposes the definition of merging conflict.The formation process of merging conflict and the types and characteristics of conflicts are then studied.On the basis of summarizing the existing conflict data collection methods,a UAV-based traffic conflict recognition system is proposed.And the influencing factors of merging conflicts are analyzed.Secondly,given the data collection strategy based on Unmanned Aerial Vehicle,video preprocessing methods such as video image processing,video stabilization,and calibration are studied.On the basis of summarizing common moving target detection algorithms and common moving target tracking algorithms,vehicle detection and tracking method based on the combination of Mask R-CNN(Region-CNN,Convolutional Neural Networks)and CSRT(Discriminative Correlation Filter with Channel and Spatial Reliability Tracking)is proposed.Combined with the calculation method of traffic conflict indicators,a comprehensive traffic conflict automatic recognition method based on Unmanned Aerial Vehicle is developed.And the application of the traffic conflict identification method was demonstrated by taking the Maqun interchange interchange area as the survey object.Thirdly,in order to analyze the merging conflict from the perspective of micro-driving behavior,the study analyzes the merging behavior from the aspects of merging execution position,merging duration,and merging speed.Then,the conflict risks related to merging decision-making behavior are summarized.Furthermore,a conflict risk prediction framework that considers the driver’s merging behavior is proposed.A multi-level logistic regression model with random parameters is used to predict the merge decision-making behavior,and then the conflict risk calculation model is used to calculate the conflict risk of each decision point.The results of the study show that the speed of the merging vehicle,the driving ability of the merging driver,and the merging execution position have a significant impact on the merging conflict.The model can not only calculate the merging conflict risk of a single merging vehicle at any time during the merging process,but also calculate the total merging conflict risk at the merge area,and then can be applied in traffic safety assessment and dangerous traffic state monitoring.Fourthly,for the problem of sample association,two modeling methods are proposed.One is to construct a joint conflict prediction model that takes into account the merging duration and the rate of merging conflicts,and aims to explore the complex relationship between merging behaviors and merging conflicts;the survival analysis model and the Tobit model are used to predict the merging duration and the merging conflict rate,respectively.The model results show that there is a coupling effect between merging behaviors and merging conflict risk.The other is to construct a merging conflict prediction model based on Bayesian networks,which aims to explore the correlation and uncertainty between influencing factors.The results of the model show that different discretization standards have a significant impact on the accuracy of conflict prediction,and there are interactions among key factors.Based on the Bayesian network model,a method for identifying traffic conflict chains is also proposed.Finally,a model strategy was proposed to predict the conflict risk of merging vehicles which could make sure that on-ramp drivers are aware of potential risks in advance.Three models(i.e.,binary logistic regression,multinomial logistic regression,and nested logit models)were developed and compared.The implementation of the proposed prediction model for merging assistance system is designed.It can have implications for the design of the merging assistance system for helping drivers make safe merging decisions,and thus enhancing the safety of the interchange merging area.