On Adaptive Optimal Control For Multi-agent Systems Via Integral Reinforcement Learning

With the expanding scale of industrial systems and the growing complexity of control tasks,elevated system performance is demanded.In recent years,the optimal control of MASs has garnered attention as a research hotspot.Should the output of the engineering system surpass the established limit of constraint,its control efficacy,stability,and safety will be severely compromised.Concurrently,there usually exists unmodeled dynamics in actual complex systems,leading to an adverse impact on the system’s performance and stability.Currently,notable advancements have been accomplished in the research of optimal control for MASs.However,the existing literature is limited in its consideration of unmodeled dynamics and output restrictions.Consequently,this thesis advances several adaptive optimal control methods for MASs with unmodeled dynamics and output restrictions.The main work and innovation of the thesis are enumerated as follows:Firstly,an optimal consensus control strategy is proposed for uncertain strict-feedback nonlinear MASs with time-varying output constraints and unmodeled dynamics.One-to-one nonlinear mapping is employed to transform constrained systems into equivalent unrestricted systems,while a dynamical signal is implemented to handle unmodeled dynamics.The feedforward controller is designed based on DSC with the introduction of error compensating signals.Additionally,the optimal feedback controller is generated through the application of ADP and IRL techniques,using NNs to approximate relevant cost functions online with designed weight updating laws.Theoretical analysis shows that designed controller not only ensures that the closed-loop system composed of all followers is SGUUB,while simultaneously maintaining the outputs within the provided time-varying constraint sets,but also guarantees minimization of the cost functions.A simulation example verifies the feasibility of the developed control algorithm.Secondly,an optimal containment control method is proposed for uncertain strict-feedback nonlinear MASs with time-varying output constraints and unmodeled dynamics.A new type iBLF is utilized to handle output constraints.A dynamical signal is applied to deal with unmodeled dynamics.The DSC approach is used to design feedforward controller.The optimal feedback controller is constructed by applying ADP and IRL techniques in which NNs are utilized to approximate the relevant cost functions online with established weight updating laws.By theoretical analysis,the outputs of all followers converge to the convex hull spanned by all leaders,and the closed-loop control system composed of the whole followers is proved to be SGUUB.In the meantime,the outputs maintain in the provided constraint sets and cost functions achieve minimization.A simulation example is supplied to illustrate the feasibility of the advanced approach.Thirdly,an optimal containment control technique is proposed for MIMO block structure MASs with time-varying output restrictions and unmodeled dynamics.The iBLF is employed to deal with output restrictions,and a dynamical signal is applied to handle unmodeled dynamics.The feedforward controller is devised using the DSC method and the error compensation signals are introduced from the second step.The optimal feedback controller is designed by introducing the idea of IRL in ADP method,utilizing gradient descent method to update the weights of NNs.Theoretical analysis demonstrates that the outputs of all followers converge to the leaders’ convex hull,and the closed-loop system composed of all followers is SGUUB.Meanwhile,the outputs remain within the specified constraint sets and the cost functions attain minimum.A simulation result validates the effectiveness of the proposed method.