Research On Autonomous Driving Situation Awareness And Risk Assessment In High-speed Mixed Confluence Environment

Autonomous driving in the high-speed mixed traffic convergence environment faces complex scenarios of dynamic interaction of multiple traffic participants.The uncertainty of the driving intention of surrounding vehicles and their trajectory brings serious challenges to autonomous driving decision-making planning.Therefore,the research on autonomous driving situation awareness and risk assessment based on the recognition of driving intention of surrounding vehicles and its long-term trajectory prediction has important theoretical value for improving the level of "humanization" decision-making and planning of autonomous driving in complex environments.Based on some of the contents of the National Natural Science Foundation of China project "Research on Autonomous Driving Situation Awareness and Humanized Robust Decision Planning in High-speed Mixed Convergence Environment"(52072054),this thesis studies the situation awareness and risk assessment of autonomous driving in highspeed mixed traffic convergence scenarios.The main research contents are as follows:(1)Recognition of driving intent of surrounding vehicles in a high-speed mixed confluence environment.By analyzing the interactive game behavior between the car and the vehicle,the dynamic space-time graph of the interactive game between the identified vehicle and its surrounding vehicles is constructed,and the graph neural network of Graph Sample and Aggregate(Graph SAGE)is used to reason about the space-time graph,and a driving intention recognition model of surrounding vehicles based on the Graph SAGE graph neural network is established,and the Softmax function is used to output straight driving,lane change to the left,The probability of three types of driving intentions to change lanes to the right and the model is trained and verified by the NGSIM(Next Generation Simulation)dataset and the accuracy of driving intention recognition.(2)Prediction of surrounding vehicle trajectory based on gated circulation unit and attention mechanism.Based on Gated Recurrent Unit(GRU)and Attention Mechanism(ATT),a trajectory prediction model for ATT-GRU encoding and decoding was constructed.The encoder encodes the historical trajectory of the predicted vehicle and its interactive vehicle,deduces its potential motion information,and integrates the driving intent recognition information,and then predicts the vehicle trajectory through the decoder.In the process of coding,the attention mechanism is introduced to explore the dependence of the vehicle’s historical trajectory on time.After the verification of NGSIM data,the driving intention information of the predicted vehicle and the attention to time can be fused to effectively improve the prediction accuracy of the trajectory.(3)Autonomous driving situation awareness and risk assessment in mixed confluence environment.Through the comprehensive analysis of various risk factors in "human-vehicle-road-environment",the gray correlation analysis method is used to determine the degree of influence of different factors on driving risk,and based on the artificial potential field theory,a driving safety risk field evaluation model considering the risks of vehicle speed,vehicle structure,dynamic traffic flow,weekly vehicle trajectory,road surface,weather environment and other risks is established,and the situation perception and risk assessment model are verified by using NGSIM convergence scenario data.The results show that the model can accurately evaluate the driving safety risk of autonomous driving in the mixed confluence environment.