Event-triggered Distributed Cooperative Identification And Learning Control

In recent years,event-triggered communication as an alternative of continuous or traditional periodic communication has received major attention in the field of multi-agent systems.In general,in order to reach the consensus or coordination control of multi-agent systems,it requires that all agents in a communication network exchange dynamical information with their neighboring agents,such as their system states.Continuous communication as an ideal communication method is widely used in many existing works,i.e.,all agents need to continuously access to the states or other information of neighboring agents.That may be impossible in many applications.It is well known that digital communication networks are often used in practice and data are received and transmitted in the form of a packet because digital signals have many advantages,such as less noise interference,robustness,signal stability,etc.However,traditional periodic communication over a digital network may not the optimal scheduling,which causes wastes of energy,bandwidth and computation resources.Especially for wireless networks,those devices are often battery-operated and energy efficiency is a critical design consideration.The general research on event-triggered communication is to precisely determine when the communication events should occur and when controllers should be updated to improve control efficiency.This dissertation is mainly concerned with event-triggered distributed cooperative identification and learning control of multi-agent systems.The communication method in the most works on distributed cooperative learning is continuous,which is not possible for practice.The event-triggered strategies make that neural weights are transmitted and received in an intermittent way.The main contributions of this dissertation are stated as follows.Firstly,the cooperative system identification of multi-agent systems is studied.A neural network(NN)based distributed cooperative identification scheme with event-triggered communication is proposed for a group of coupled identical nonlinear systems.Different from the existing event-triggered strategies in dynamical control systems,the proposed event-triggered condition,relating to their NN weights rather than the state,can still guarantee that the NN weights of all identifiers converge to a small neighborhood of their optimal values along the union trajectory.The overall computation and communication messages are reduced,and meanwhile,the approximation performance on an union domain is still maintained.It is further proved that there exists a positive minimum inter-event interval and Zeno behavior can be avoided.Secondly,the distributed cooperative learning control for a group of uncertain nonlinear systems is investigated in chapter 4.An event-triggered condition based on NN weights is proposed to overcome the disadvantages of continuous communication.Each agent intermittently broadcasts its NN weight estimation to its neighboring agents under the proposed event-triggered condition.This relaxes the requirement that all agents continuously access to the weight information of their neighbors.Then,we design an NN weight update law in the event-triggered context.The tracking performance of the adaptive NN control systems are still maintained well,and all radial basis function(RBF)NNs with a small approximation error is obtained on an union domain.It is illustrated that the past experience obtained in the event-triggered context can be used to improve the control performance.Furthermore,a strictly positive lower bound on the inter-event intervals is also guaranteed.This means that all agents in a network do not exhibit Zeno behavior.Finally,the event-triggered cooperative learning method for output feedback control of multi-agent systems is studied in chapter 5.An event-triggered communication scheme for distributed cooperative learning is proposed to avoid continuous communication.For each agent,its trigger function is only dependent on the weight estimate about itself rather than its neighboring agents.Each agent exchanges the NN weight information with its neighboring agents in a discrete manner during the control process,which allows for more realistic implementation.The proposed scheme effectively reduces the communication load.An event-based adaptive distributed cooperative learning(DCL)law for output feedback NN control is designed.The proposed learning law still can guarantee that the weights of RBF NNs converge to a small neighborhood of their optimal values,and,meanwhile,the system orbits can track the reference orbits with a small error.It is proved that the inter-event times are lower bounded by a positive constant to avoid the accumulation of events.