Safety Modeling And Management Strategy Simulations And Evaluations For Motorized And Non-motorized Mixed Traffic

In recent years,the non-motorized vehicles have played an important role in meeting the rapid demand of short-distance travel.The main traffic structure of our country is formed from the non-motorized vehicle and the motorized vehicle.The motorized vehicle solves the problem of medium and long-distance travel.The development of non-motorized vehicle traffic brings great convenience for daily traveling,however,it also exposes a series of problems of road traffic status in our country.For example,the width of the non-motorized vehicle lane is not enough.The separation form of the non-motorized vehicle lane and the motorized vehicle lane is not clear.Many non-motorized vehicles ride into the motorized lane leading to a complicated mixed traffic condition.The mixed traffic reduces the efficiency of traffic operation and easily leads to traffic accidents.Therefore,this study conducted research on the “safety modeling and management strategy evaluations for motorized and non-motorized mixed traffic” based on the National Key Research and Development Program of China “Key Technologies and Equipment for Emergency Treatment of Road Traffic Accidents Under Multi-Source Information Environment”(No.2018YFE0102700).This study focuses on the scene of motorized vehicles and nonmotorized vehicles mixed traffic and starts from the level of microscopic interaction behavior in the mixed driving environment.This study was carried out from three aspects,data acquisition,micro-behavior modeling and simulation,and road control and evaluation.First,the research scenario of mixed traffic was defined,and the separated types between motorized and non-motorized vehicle lanes and interaction behavior types were described.The observation sites were selected based on the interaction behavior types.The data selection method was proposed,which was based on computer vision technique and manual selection.The computer vision technique provides trajectory data by target detection and tracking,coordinate transformation,background generation,and trajectory noise smoothing.This method can give accurate trajectory data.Second,the interaction behavior types were divided in detail based on the types of vehicle and lane separation.The characteristics of each type of interaction were analyzed.The typical interaction behaviors were analyzed in detail,which include the overtaking behavior between non-motorized vehicles,friction interaction,and block interaction.The Kaplan-Meier model was used to explore the duration of non-motorized overtaking behavior.The Bayesian Logistic model was used to study the influence factors of the non-motorized vehicle to motorized vehicle interaction behavior.The Bayesian Logistic model considered individual heterogeneity and observation site heterogeneity and the power of model interpretation was improved.Third,the mechanism of the motorized-nonmotorized vehicle interaction behavior was studied.The critical lateral distance was used as the indicator to judge the occurrence of the interaction behavior.The fitting distribution of critical lateral distance was explored,and a fully parametric accelerated failure time duration model was built to study the interaction behavior.The critical lateral distance duration model was developed based on the classic time duration model,and the critical lateral distance was treated as the analogy of the time variable.In addition,the parametric accelerated time duration model considered the unobserved heterogeneity to improve the model’s fitting accuracy and the power of data interpretation.The model investigated the influence of several factors on the interaction behavior such as speed,speed difference,yaw rate,yaw rate difference,acceleration,traffic volume,driving position,vehicle type,and load state.Fourth,the relationship between conflict behavior and interaction behavior was clarified.The conflict behaviors were extracted by using traffic conflict technology,and the agent-based model was used to simulate the conflict behavior trajectories.In the modeling framework,the reward functions were obtained by the inverse reinforcement learning algorithm.The recovered reward functions provide the behavioral preferences of the motorized and non-motorized vehicles.After that,the strategies of motorized and non-motorized vehicles were estimated by using the deep learning algorithm,and their trajectories were predicted.The simulation accuracy of the algorithm was judged.Both the multi-agent model and single-agent model were developed in each step of the trajectory simulation,and the simulation results and accuracies of the two models were compared.The multi-agent model was superior to the single-agent model in the trajectory prediction accuracy and capturing the intelligence and rationality of road users.In the end,the conflict time proximity indicators(TTC and PET)were calculated from the predicted trajectories.The predicted indicators had high correlations with the actual ones.Finally,five control strategies were proposed based on the characteristics of interaction and conflict behaviors.The VISSIM system was employed to set up mixed traffic simulation scenarios for extracting efficiency indices(e.g.,speed and delay)and vehicle trajectories under different control strategies.The emulated vehicle trajectories were input into SSAM software to calculate safety indices(e.g.,conflict number)of mixed traffic under different control strategies.Efficiency and safety indices before and after the implementation of control strategies were respectively compared to the efficiency and safety influences of control strategies on mixed traffic from a macroscopic perspective.Moreover,the simulated efficiency indices were considered as the input variables of the conflict behavior simulation model.A series of conflict behavior simulations were conducted to estimate safety indices under different control strategies,and then the impact of control strategies on mixed traffic safety was analyzed from a microscopic perspective.In this study,the characteristics and occurrence mechanisms of various micro-interaction behaviors in the mixed traffic environment were explored.The influence factors on the probability of interaction behaviors were quantified.The results of the study enrich the modeling and simulation system of mixed traffic research on the theoretical level and provide theoretical and data references for the construction of mixed traffic facilities.They also provide a method for improving the operation efficiency and safety of the mixed traffic environment.