Data-driven Platoon Control Of Vehicles In A Connected Environment

Vehicle platoon control has great potential in solving some pivotal problems of today's road transportation system,such as alleviating traffic congestion,improving road safety,energy saving and emission reduction.It will bring quite significant economic and social benefits.This paper studies the problems of communication loss,inability to accurately model vehicles,and tracking and switching of platoon vehicles in the Cooperative Adaptive Cruise Control(CACC)system in the longitudinal platoon of vehicles.The main research contents are as follows:(1)Usually the dynamic parameters of vehicles in a longitudinal platoon system are unknown,and traditional control methods cannot achieve optimal control without obtaining an accurate model of the system.To this end,the paper firstly presents a centralized vehicle longitudinal platoon model expression,and analyzes the influence of its dynamic parameter changes on the vehicle trajectory.Furthermore,an adaptive dynamic programming centralized control method based on reinforcement learning is proposed,which only uses the input and output data to realize the optimal control of the vehicle platoon.Compared with the adaptive dynamic programming control method based on neural network,the proposed method can realize online learning of system knowledge and real-time control,and does not need to input a large number of samples for training in advance.(2)When a vehicle in the longitudinal vehicle platoon performs a lane change and departure action or a vehicle outside the platoon tries to perform a merging action,the following vehicle needs to switch its tracking target.To this end,the paper uses the idea of virtual structure,and comprehensively considers the driving state of the tracked vehicle before and after switching the target to construct a virtual vehicle to achieve smooth tracking of the new target.The method overcomes the large tracking error caused by the sudden switching of the tracking target,and reduces the acceleration of the vehicle during the switching process.Simulation verifies the smooth performance of the proposed method under vehicle tracking target switching.(3)Affected by the state of network communication,there is a frame loss phenomenon in the information transmission in the vehicle network.Under the assumption that there is no sudden change in the speed of the fleet,a vehicle state estimation algorithm based on adaptive unscented Kalman filtering is proposed to estimate the vehicle speed and distance in real time to compensate for the lost frame data of communication.The simulation analysis compares the influence of the communication frame loss data compensation method on the vehicle queue tracking performance under different communication packet loss rates,and verifies the effectiveness of the proposed vehicle state estimation and compensation method.