Research On Vehicle Cooperative Control For Enhanced Safety Under Vehicle-road Coordination

In urban traffic environment,vehicles need to follow and change lanes frequently,due to lack of information and wrong decision-making,human drivers are prone to bring safety hazards to traffic and reduce traffic efficiency.The vehicle-road coordination system can expand the perception range of connected and automated vehicles through vehicleroad communication,and can also assist or take over vehicle control through the on-board control model to improve decision-making efficiency.The safety level of vehicle control is closely related to passenger safety,with the establishment and improvement of 3 GPP,ITU-R standards,higher requirements have been made for the driving safety of the vehicle-road coordination system.From theory to practical application,the vehicle-road coordination system still needs to solve key vehicle safety issues.For the safety-enhanced vehicle cooperative control under vehicle-road coordination,this thesis studies typical applications in 3GPP standard such as vehicle platoon and vehicle lane changing,respectively.For vehicle platoon application,the existing vehicle platoon control model makes decisions based on an ideal vehicle following model,which has potential safety hazards due to the inevitable perception error of onboard sensors.To make vehicle platoons pass through intersections safely and efficiently,this thesis proposes a vehicle platoon control model based on vehicle-road coordination.Firstly,according to the signal information at the intersection and the estimated arrival time of the vehicles,a dynamic platoon division scheme is proposed to realize the role switching of the leader vehicle and the follower vehicle in the platoon.Secondly,for the longitudinal speed control model of the leader vehicle,the triangular linear algorithm is used to optimize the speed of the vehicle based on the signal information at the intersection,which can slow down the speed change rate of the platoon and reduce the congestion at the intersection;Aiming at the longitudinal speed control model of the follower vehicle,a vehicle following model considered perception error is proposed to dynamically describe the speed control of the follower vehicle under the influence of the error,then the fusion of differential global positioning perception data is used to make up for the defect of single vehicle perception and reduce the safety hazards of follower vehicles.Finally,the effectiveness of the scheme in improving intersection efficiency and traffic safety is verified by simulation.For vehicle lane changing application,the artificially formulated rulechanging model is difficult to deal with random traffic environment,which is easy to cause safety risks,while the data-based deep reinforcement learning can train a complete model through the interaction with the environment.However,the trial-and-error nature of traditional reinforcement learning can easily lead to vehicle conflicts during training,which is not conducive to the practical application of the algorithm.This thesis proposes a vehicle lane changing training algorithm based on safe deep reinforcement learning,and implements risk identification and safe action search by introducing a safety module into traditional deep reinforcement learning.Firstly,according to the vehicle position and action,a vehicle position convex estimation algorithm is proposed to predict the future collision risk of the vehicle;Secondly,for the risk state,the pre-exploration based on the domain randomization virtual environment is adopted to avoid the vehicle collision in the main environment,and accumulate virtual risk samples to accelerate the learning of hazardous actions;Then,a Monte Carlo search tree algorithm is introduced to search for safe actions in the main environment.Finally,the effectiveness of this algorithm in avoiding vehicle conflicts and improving training efficiency is verified through simulation experiments,and the training model meets the delay requirements of vehicle control under the 3 GPP standard.