Cruise Control For Vehicles Platoon Based On Deep Reinforcement Learning

Connected autonomous technology is one of the core technologies in the intelligent transportation systems(ITS).Connected autonomous vehicles(CAVs)mainly drive longitudinally along the road on the highway.The goal of CAVs formation control is to design a car-following optimization algorithm that enables autonomous vehicles to travel stably in longitudinal formations.Therefore,this article will design deep reinforcement learning algorithms for driverless vehicle formations and hybrid driving vehicle queues to achieve direct control from input to output,and further solve the control problems in the field of autonomous driving.The main research contents of this paper are as follows:First of all,aiming at the various traffic conditions and limited experimental equipment conditions,an intelligent transportation experiment platform was built,which is mainly composed of ISRV intelligent vehicles,visual information positioning systems,and wireless communication systems.Independently designed hardware and software.In order to ensure that the designed deep reinforcement learning algorithm can be effectively implemented in the vehicle formation control system,it provides a guarantee for the experimental verification of the following theoretical algorithm.Secondly,considering the car-following problem of unmanned vehicle formation,a queue control method based on deep reinforcement learning algorithm is proposed.Using vehicle-to-vehicle communication,a dynamic model of the car-following unmanned vehicle formation is established.Furthermore,the Markov decision design is carried out on the car-following behavior of the unmanned vehicle formation,and the Multi agent Q learning-deep deterministic policy gradient algorithm based on multi-agent Q-learning is proposed.It can achieve complete data-driven and self-learning,without relying on any predictions,predefined rules or even previous human knowledge,so that the vehicles in the formation can form a predetermined formation and maintain the ideal platoon speed,while ensuring the stability of the entire formation system.The proposed car-following control algorithm is verified by simulation,and the practicability of the algorithm is verified by experiments.Finally,for the longitudinal formation of mixed vehicles composed of connected autonomous vehicles and human-driven vehicles,fully considering the parameter uncertainty of manned vehicles,a queue control strategy based on deep reinforcement learning is designed.For the longitudinal formation of mixed vehicles,a general car-following mathematical model was established,and the position of the two types of vehicles can be changed freely.Further analyze and design the Multi network time-deterministic policy gradient algorithm based on the update cycle of the multi-agent value network,which realizes the control of the vehicle.High-level tracking control accuracy and stable learning simplifies the optimization solution process and speeds up the convergence speed.Then proved the convergence and non-bias of the algorithm,showing the superiority of the algorithm.And through simulation and experiment,the effectiveness of the proposed car-following optimization control algorithm is further verified.