Study On Lane Change Decision By Stackelberg Game Considering Driving Styles Of Interacting Vehicles

With the continuous development of intelligent driving technology and the advancement of national policies and regulations,autonomous vehicles,as an important part of the future intelligent transportation system,have entered the stage of field testing and preliminary commercialization.In the hybrid driving scenario where the autonomous vehicle and human-driving vehicle are driving together,in order to meet the safety,efficiency and comfort requirements of the occupants of the unmanned vehicle,in addition to driving path planning and obstacle avoidance,the interaction behavior decision-making between the autonomous vehicle and other vehicles should be studied.Based on driving behavior data,lane changing interaction behavior is modeled by using the Stackelberg game model considering driving styles of interacting vehicles,and lane changing decisions of autonomous driving vehicles in this scenario are studied.The main research work is as follows:(1)Data screening and analysis based on high D driving data set: In the selected driving data set of high D,the driving lane change data required in this paper is extracted through vehicle size type,operation mode and interaction form.Furthermore,the key stages of lane changing are divided to study the behavior characteristics of lane changing at different stages.In addition,the driving behavior of the vehicle behind in the target lane in the two stages of lane change and lane change trial is numerically tested to analyze the influence of lane change trial behavior of the vehicle in front on the vehicle behind in the target lane.(2)The classification and recognition of driving style of lagging vehicle in the target lane: The driving characteristic quantity of the vehicle behind in the target lane during the lane change trial period was extracted,and standardized treatment was carried out after screening outliers.Principal component analysis(PCA)is used to reduce the dimensionality of high-dimensional eigenvectors to low-dimensional PCA vectors.The principal component eigenvectors are classified into three types: cautious,ordinary and aggressive by the K-means algorithm.Three supervised learning methods,namely BP neural network,LVQ neural network and support vector machine,are used to identify and learn the relationship between feature principal components and driving style.By comparing the recognition effects of the three algorithms,the support vector machine algorithm after parameter optimization is finally confirmed as the driving style recognition algorithm.(3)Considering the driving style of the vehicle behind the target lane,a lane change game decision model is established: The actual lane changing interaction behavior is explained by the Stackelberg game theory,and the game types,participants and strategy sets in the lane changing stage are defined.The feasible space of lane changing and the starting and ending conditions of the game were established by modeling lane changing trajectory,and the game payoff function of each decision was determined by analyzing driving risks and driving intentions of both parties involved in lane changing,and the optimal decision of both parties was determined by using Stackelberg game deduction.The simulation scenario was established to solve the established Stackelberg game model,and the safety of lane change was analyzed and verified.Finally,by extending the simulation scenarios,the optimal payoff and decision difference under different initial intervals are studied,which provides a reference for the decision-making of the autonomous vehicle in different lane changing scenarios.