Research On The Expressway Vehicle Driving Decision-Making Model Based On Segmental Full Trajectory Data

With the rapid construction and development of our country’s expressways,it is also facing the problem of high frequency of traffic accidents.Car-following and lane-changing are common decision-making behaviors in driving.How to reduce traffic accidents caused by driver’s wrong decision-making and improve road safety has become an urgent problem to be solved.With the increasing maturity of intelligent network technology,vehicles can perceive more accurate driving environment information,which provides the possibility to improve the vehicle’s autonomous or assisted decision-making capabilities.Therefore,based on the full trajectory data of the expressway section,this thesis adopts a data-driven research method to carry out related research on vehicle following and lane-changing decisions from a micro perspective.The main work is as follows:First of all,aerial vehicle trajectory dataset.Use a drone to take high-altitude aerial shots of the vehicle driving video of the highway exit ramp section,and analyze the video vehicle data through the target detection and tracking algorithm,and extract all the vehicle trajectory data in the section;based on the traffic flow theory,analyze the traffic flow in the observation area The spatio-temporal characteristics of traffic flow parameters clarify the traffic characteristics of the data;screen the vehicle following and lane changing behavior data,and mark the key points of the vehicle lane changing process according to the motion characteristics of different stages of lane changing,laying a data foundation for subsequent research.Secondly,establishment of vehicle following and free lane-changing decision-making model fully.Consider the influence of driver personality characteristics on driving decisionmaking,extract the features used to characterize the driving style of car-following and free lanechanging behavior,and conduct cluster analysis and classification identification of driving style.For car-following behavior,on the basis of considering driving style,Kalman filter is used to establish a combined car-following model that combines IDM(theory-driven)car-following model and LSTM(data-driven)to determine the car-following speed in real time to reduce A vehicle rear-end collision accident caused by a wrong decision in the process of following a vehicle.Aiming at free lane-changing behavior,combined with driving style,a lane-changing decision model is constructed using a gated neural unit(GRU)and a self-attention mechanism,and a lane-changing decision is made in real time to provide safety guarantees for vehicle lanechanging execution.Then,establishing a decision-making model for forced lane-changing vehicles.In order to explore the forced lane change behavior of vehicles in the off-ramp scene,the difference between the lane change and free lane change is compared and analyzed;considering the microscopic dynamic characteristics of the vehicle’s forced lane change behavior,the distribution law of the location of the lane change execution point and the distribution law of the lane change length are analyzed;A forced lane change decision model based on random forest is established,and the Bayesian optimization algorithm is used to determine the optimal hyperparameter space to realize the vehicle’s forced lane change decision under road constraints.Finally,a Python-Sumo joint simulation platform is established to build a simulation experiment scene;combined with the vehicle car-following and lane-changing motion characteristics,the vehicle-following and lane-changing decisions are simulated in different scenarios,and the validity of the model established in this thesis is verified from a microscopic perspective.The research in this thesis can provide technical support and decision-making basis for promoting vehicle driving safety and expressway traffic control implementation to a certain extent.