Research On Cluster Consensus Of Multi-agent Systems Based On Event-Triggered Mechanisms

In recent years,cooperative control of multi-agent systems(MASs)has attracted increasing attention from scholars in the field of control sciences.As a basic problem of cooperative control of MASs,consensus has witnessed a wide range of practical applications involving unmanned aerial vehicle(UAV)formation,multi-robot motion and so on.In practice,an MAS often needs to be divided into multiple clusters.The number of clusters,the number of agents in each cluster and the topologies may vary while there are certain changes in tasks,environments and communications.In an MAS,agents which are usually distributed spatially communicate with each other via a network.Event-triggered strategies contribute a compromise between satisfactory system performance and limited communication resources.However,different clusters have different information updating requirements due to distinctive physical environments or network conditions.Thus it is urgent to explore effective event-triggered mechanisms to distinguish the information updating frequency for agents from different clusters.Furthermore,it is often difficult for agents to acquire complete and accurate state information from neighbors,and it is technically easier to design feedback control schemes using directly measured output information than state feedback.However,how to solve the non-convex problem in output feedback control design with less cost or less conservativeness is still an opening topic.In recent years,intelligent algorithms have been favored by researchers because of their superiority in processing large amounts of data.It is worth exploring using intelligent algorithms based methods to solve the non-convex issue in terms of matrix inequalities.In this thesis,aiming at typical linear multi-agent systems,a series of novel cluster eventtriggered mechanisms,cluster consensus control protocols,and intelligent algorithms based control gain and triggering parameter co-design methods are proposed to investigate cluster consensus of MASs.The main contents of this thesis are outlined as follows:(1)Two static cluster event-triggered mechanisms are proposed,where the event triggering condition is separated into two parts: intra-cluster and outer-cluster,and several parts: intra-cluster and inter-cluster,by assigning disparate weighting matrices and threshold parameters to distinguish different information updating requirements of agents.Based on the two static cluster eventtriggered mechanisms,a cluster consensus control protocol is designed.An effective multipopulation genetic algorithm(GA)with an accumulated fitness function is proposed to solve the non-convex problem in terms of matrix inequalities caused by the coupling of output feedback control gain and variable matrices.The control gains and event triggering threshold parameters are designed jointly.Numerical experiments on a multi-satellite formation show that the control protocol based on the two static cluster event-triggered mechanisms can achieve cluster consensus of MASs,and the effectiveness of the two static cluster event-triggered mechanisms is verified by comparing with other event-triggered mechanisms.(2)A dynamic cluster event-triggered mechanism with dynamic threshold parameters based on combination measurement is proposed,where the event triggering condition is divided into two parts: intra-cluster and outer-cluster.Two dynamic threshold parameters with different initial values and different dynamic change rules are assigned.On the one hand,by introducing dynamic threshold parameters,the event triggering condition is dynamically adjusted to balance system performance and network resource saving more flexibly;on the other hand,the relative states between an agent and its neighbors at the current moment and the latest trigger moment are regarded as the error state in event triggering condition,which is more favorable to associate the state update of the agent with the whole system.Based on the proposed dynamic cluster event-triggered mechanism,control protocols are designed under fixed and Markov switching network topologies.An improved particle swarm optimization(PSO)algorithm is proposed to design the initial dynamic threshold parameters and control gains jointly.Numerical experiments on a multi-satellite formation show that cluster consensus of MASs can be achieved under the proposed dynamic cluster event-triggered mechanism and control protocols,and the superiority of this dynamic cluster event-triggered mechanism is demonstrated by comparing with other mechanisms.(3)A resilient dynamic cluster event-triggered mechanism is proposed,where resilient coefficients are introduced into the dynamic change rules of threshold parameters.By designing resilient coefficients for agents in different clusters,different parameters are selected for agents in different clusters according to actual network conditions.Thus all the clusters are classified into two parts: one owns free network and the other owns congested network.Then the accumulated errors in the event-triggered condition are divided into two parts accordingly.A well balance between desired system performance and limited network resources can be made for the whole MAS.Based on the proposed resilient dynamic cluster event-triggered mechanism,the cooperative control protocols of homogeneous and heterogeneous MASs are designed.The control gains and dynamic threshold parameters are co-designed by combining PSO with GA by introducing the crossover and mutation operation of GA into PSO.Numerical experiments on a multi-satellite formation show that the cluster consensus of homogeneous and heterogeneous MASs can be achieved under the proposed resilient dynamic cluster event-triggered mechanism and control protocols,and the effectiveness of the proposed resilient dynamic cluster event-triggered mechanism is verified by designing different resilient coefficients in diverse time interval.