Study On Man-machine Co-driving Control Switching During Lane Changing In Diverging Area Of Expressway

With the development of computer,artificial intelligence and other new technologies,it directly promotes the emergence of human-machine co-driving and even driverless cars.The diverging area of freeway is an important part of freeway and a typical bottleneck of freeway network.Statistics show that 83% of the freeway entrance and exit accidents occurred in the diverging area of the freeway.Based on the lane-changing scenario in the diversion area of the freeway,this paper analyzes the driving ability and task demand ability under the condition of human-machine co-driving,constructs the corresponding risk assessment model and capability assessment model,and studies the switching timing in the process of compulsory lane-changing by combining the capability matching theory and reinforcement learning theory.The specific research contents are as follows:(1)Build the capability assessment model of lane changing task demand in diverging area of freeway.Based on NGSIM data set,the characteristics of forced lane change in the diverging area of freeway,such as lane change execution time,vehicle speed before and after lane change and time headway in front,are analyzed.Based on fuzzy logic theory,the risk evaluation model of lane change process was established considering longitudinal acceleration and transverse velocity absolute value indexes.The evaluation index of lane changing task demand ability was selected,and the weight of each index was determined based on entropy weight method,and the assessment model of lane changing task demand ability was constructed.(2)Establish a driver’s ability assessment model.Based on normalization and PCA,a driver characteristic characterization index system was constructed,which considered the characteristics of speed,acceleration and displacement.The driver characteristics were analyzed based on clustering algorithm,and three clustering methods,K-Means++,FCM and hierarchical clustering,were analyzed and compared to solve the driver style optimization parameters.A risk assessment model considering driver characteristics is proposed and a driver capability assessment model is established.(3)Establish a driving system capability assessment model.Based on the trajectory data preprocessed by NGSIM,vehicle position coordinates,velocity and acceleration were selected to extract trajectory feature vectors,and lane-change trajectory prediction model was established based on LSTM,and lane-change trajectory tracking was carried out based on MPC.Based on the simulation results of lane changing model of automatic driving system,the lane changing ability of driving system is evaluated by fuzzy logic algorithm.(4)Proposed the human-machine co-driving control switching strategy based on TD3 algorithm.Based on the ideal solution approximation method(TOPSIS),the data processing of lane-changing task requirement ability,driver’s ability level and driving system’s ability level were carried out.The TD3 algorithm was selected as the decision learning algorithm of the model,and reasonable strategy learning objectives were set for the model.By strengthening the reward and punishment mechanism of learning,the process of control control gradually switching from the driving system to the driver when the vehicle encounters risks was realized,providing theoretical basis for personalized control control switching strategy of human-machine co-driving.