Research On The Consensus Control Algorithms For Multi-agent Systems Based On Adaptive Dynamic Programming

In recent years,with the rapid development of sensor technology and network technology,simple single agent system has been unable to meet the requirements of the current increasingly complex system,so more flexible and complex multi-agent system emerged.Research direction of the multi-agent control problem can be divided into flocking control,formation control,cluster control,consensus control and containment control,etc.Most of the multi-agent group control problems are based on the consensus theory,therefore the consensus problem of the multi-agent system control has become the current hot research topic in the field of multi-agent control.Adaptive dynamic programming is a popular method to solve optimal control problems.This paper mainly studies online optimal consensus control algorithm of multi-agent systems based on adaptive dynamic programming.The main research contents include the following aspects:Firstly,aiming at the problem of online optimal consensus control of discrete-time multi-agent systems,the problems of on-policy adaptive dynamic programming algorithm are analyzed,and an online optimal consistency control algorithm based on off-policy adaptive dynamic programming is proposed.In this algorithm,by introducing a new Actor neural network,the policy obtained after policy promotion is directly applied to its corresponding state,which overcomes the problem that the policy is inconsistent with the actual state after the policy promotion of the traditional on-policy algorithm,and improves the convergence efficiency of the algorithm.Secondly,when solving the problem of online optimal consensus control for discrete-time multi-agent systems,both on-policy and off-policy adaptive dynamic programming algorithms need a lot of computation and communication bandwidth.In order to solve this problem,an adaptive dynamic programming optimal consistency control algorithm based on event-triggerred scheme is proposed in this paper.In this algorithm,the event-triggerred scheme is used to replace the time-triggerred scheme of the traditional algorithm,so as to avoid the waste of some unnecessary computing resources.Each agent designs its own event triggering condition to avoid that the whole discrete time multi-agent system will not be triggered at the same time,so as to reduce the communication load of the whole system at the same time.Thirdly,in order to verify the correctness of the proposed algorithm,a multi-agent algorithm verification platform based on the quadrotor UAV cluster system is designed and built in this paper.In this platform,communication between u AVs and between UAVs and ground station system is carried out through robotic operating system(ROS).Algorithms are deployed on the airborne raspberry PI of UAVs,and the operation of UAVs is controlled by algorithms in airborne Raspberry PI.