Research On Distributed Consensus Control And Optimization Of Multi-agent Systems

Recently,distributed cooperative control of multi-agent systems has been received more and more attention from multi-disciplinary researchers because of its wide applications to co-operative attitude control of satellites,cooperative control of unmanned aerial vehicles,coor-dinated dispatch of transportation system and many other practical engineering applications.As a basic cluster behavior of distributed cooperative control,distributed consensus control refers to design a distributed control protocol based on local information that enables all the agents to reach an agreement regarding a certain quantity of interest.In practical applications,the designed controller not only needs to ensure the stability of the system but also guarantees the performance index of the system optimal.Therefore,this paper studies the distributed consensus problem of multi-agent systems from two aspects:control and optimization.The main contents of this paper are summarized as follows:Finite-time distributed event-triggered consensus control of multi-agent systems is stud-ied.In order to alleviate the system's communication pressure and reduce the number of cal-culations for the system,we introduce a new data-sampled method—event-triggered control mechanism.Compared with the traditional periodic sampling,one task is implemented only when a pre-defined event is satisfied in the event-triggered control.We propose a finite-time distributed event-triggered consensus control strategy for fixed and time-varying two differ-ent communication network topologies,respectively.Subsequently,the Lyapunov functional method and the inequality zoom technique is utilized to demonstrate that all the agents can reach consensus in a finite time.The distributed optimal consensus control problem for linear continuous-time multi-agent systems with time varying input delay is investigated.While stable controllers are inspiring,system performance is generally preferred for multi-agent systems.To research optimal consensus control problem of input-delayed multi-agent systems,we firstly convert optimal consensus problem of linear continuous-time input-delayed multi-agent systems in-to that of linear discrete-time delay-free multi-agent systems based on model transformation and performance equivalent.Then,according to Bellman optimization principle and adaptive dynamic programming technique,we design a distributed optimal consensus control policy for delay-free multi-agent systems.In order to implement adaptive dynamic programming method,we introduce the critic-actor neural networks structure to approximate the optimal value functions and the optimal control policies.Data-driven distributed optimal consensus control for unknown input-delayed multi-agent systems is proposed.For the case of the unknown system model,we propose a data-driven-based distributed optimal control policy for each agent.Compared with the traditional model identification method,data-driven control scheme not only eliminates the errors gener-ated by the model identification,but also reduces computation and the communication trans-mission.In order to simplify the analysis,we employ the model reduction method to trans-form optimal consensus problem of input-delayed multi-agent systems into that of delay-free multi-agent systems.Then,data-driven-based adaptive dynamic programming technique is used to solve optimal consensus control of delay-free multi-agent systems.Distributed optimal event-triggered consensus control for nonlinear multi-agent systems using adaptive dynamic programming is analyzed.Combining event-triggered control with adaptive dynamic programming technique,we propose a new method to address the opti-mization problem-event-triggered adaptive dynamic programming.In the proposed event-triggered adaptive dynamic programming structure,the control policies and weight estima-tions are updated only when the designed event-triggered conditions are violated.Therefore,compared with the traditional time-driven adaptive dynamic programming,this method can reduce computational burden and communication rate so that computation and communi-cation resources are saved largely.Then we introduce fuzzy hyperbolic model-based critic neural network framework to approximate the optimal value functions and help calculate the optimal control policies.Compared with the dual critic-actor neural networks structure,it is more reasonable to utilize single critic network framework for multi-agent systems due to simpler network structure and fewer number of training.Finally,we make a simple summary of our work and discuss the research work that will be accomplished in the future.