Research On Quantized Learning Consensus For Several Types Of Multi-Agent Systems

Multi-agent system is a distributed independent system composed by multiple agents,which can solve the problems that a single agent can not complete and improve the efficiency and robustness.With the rapid development of network and communication technology,the networked control system has become a hot research.Due to the addition of the network,the networked control system will be affected by some network factors,such as network communication bandwidth and network load capacity constraints,so there will be some net-worked problems,such as packet loss,delay and so on,which are the inevitable and major problems in the study of networked control systems.Therefore,in order to make full use of the limited communication bandwidth and reduce the information load effectively,we will introduce quantization into the design of the consensus protocol of the multi-agent system,and the iterative learning control algorithm is combined to discuss the tracking problem-s of the leader-following multi-agent systems,and the conclusion of accurate tracking is obtained.The main works of this dissertation are:1.The consensus problem of leader-following multi-agent systems by using the robust learn-ing control approach is investigates,and a visible distributed quantized protocol is also given to update the dynamic systems.Because of the existence of the quantizer,even linear sys-tems turn to be nonlinear ones.A robust control scheme with the help of Lyapunov direct method is utilized to overcome the difficulty.Compared with the systems without quan-tization,the new robust learning control guarantees the asymptotic tracking property.For quantized dynamic systems with nonlinearities and uncertainties,the algorithm reflects the quantized protocol of good robustness.In addition,to reduce the communication resources further,the networked problem with event-trigger and without event-trigger are considered in the design of the controller respectively.What's more,the criteria for the asymptotical convergence analysis and robust controller of quantized multi-agent system are established by utilizing Lyapunov function.2.In order to achieve the requirement of utilizing the digital channel more effective,this dis-sertation investigates the consensus problem of the quantized iterative learning control ap-proach for the linear and nonlinear uncertain continuous-time multi-agent system,and gives a visible distributed quantized protocol to update the dynamic systems with Sigma-Delta(??)quantizer which has a finite number of quantization level to realize perfect tracking for the multi-agent system.Due to the exist of quantization and other networked problems,a robust compensation learning control scheme with the help of the Lyapunov direct method and the non-smooth analysis is utilized to eliminate the quantization errors and system un-certainties,and the asymptotical convergence analysis can be established based on randomly small number of quantization bits,even merely one bit of quantization information exchange between each pair of adjacent agents can realize target tracking.Moreover,to obtain the ex-act tracking property as iterations tend to infinity,we introduce a saturation function and a special piecewise function into the control inputs and make the problem of networked control more comprehensive.3.A problem of quantized iterative learning control for continuous-time multi-agent sys-tems with finite-leveled??quantization and random packet losses is firstly proposed in this dissertation.To realize the digital communication between signals and utilize limited communication bandwidth effectively,we introduce the??quantizer with limited commu-nication data rate(quantization bits)into the control field and for the design of the quantized learning control in our dissertation.In addition,the packet losses are also firstly considered into the quantized iterative learning control,which makes the controller more close to the practical engineering applications.Since the nonlinearity and randomness introduced by the quantization and packet losses,a decreasing learning gain is utilized with the help of the non-smooth analysis and mathematical expectation for the analysis of convergence.Accu-rate tracking in the sense of expectation can be obtained based on randomly small number of quantization bits,even merely one bit of quantization information.4.This dissertation proposed a quantized iterative learning control for continuous-time non-linear multi-agent systems with logarithmic quantization and arbitrary initial states.By introducing the quantization,digital communication between signals is realized and the re-quirement of utilizing the digital channel more effective is also achieved.In order to elimi-nate the initial state errors introduced by the assumption of any initial value for every agent and to avoid the complexity for the common method of initial-state learning,a new method of control inputs compensation is introduced in this dissertation and the accurate tracking over the entire time period can be obtained asymptotically by the analysis of convergence with the help of non-smooth analysis and the principle of compressed mapping.5.This dissertation develops a framework for treating multi-agent consensus problems us-ing quantized control.Unlike the previous static quantizer,such as uniform quantizer or logarithm quantizer,this dissertation proposes a new quantized consensus protocol through dynamic??quantizer for multi-agent dynamical systems,which makes the system achieve asymptotic consensus with finite bits and converge to the average of the initial states.More-over,the sufficient condition is given.Compared with the asymmetrically and symmetrically quantizer,??quantizer overcomes the disadvantages of the no memory and the existence of the steady state error about the static quantizer and the need of quantification with unlimited bits of information,which reflects its advantage.6.This dissertation investigates the consensus problem for event-triggered multi-agent sys-tems subject to external disturbances,time delays and packet losses as well as??quantizer who has a finite number of quantization level simultaneously.Firstly,we transform this problem into a robust H_?control by defining an appropriate controlled output.Secondly,we give a criteria to judge the consensusability of asynchronous event-triggered MAS and present a sufficient condition in terms of matrix inequalities to get the state feedback con-troller's parameters.Moreover,the maximal allowable number of successive packet losses is also given,and the algorithm designed here has made considerable contributions to save communication resources.