Research On Decentralized Adaptive Critic Control For Nonlinear Interconnected Systems

Adaptive critic control,which is also known as adaptive dynamic programming(ADP),combines the ideas of the optimal control,neural networks and reinforcement learning.Adaptive critical control is a near-optimal method in the optimal control field and can effectively overcome the traditional “curse of dimensionality”.With the development of economics,societies,and industries,the complex nonlinear interconnected systems are required to be optimized and controlled.These complex systems usually subject to state time-delays,unknown system dynamics,strong nonlinearities,interconnections,input constrains,etc.These limitations result from potential challenge for traditional control theory and methods.Decentralized control is an efficient control method for complex interconnected systems.It aims to find a series of local control policies for interconnected subsystems to solve optimal control problems.Since there is no information exchange among subsystems,it does not only reduces the computational burden but also improves the scalability of the developed control systems.Therefore,the research on adaptive-critic-based decentralized control for nonlinear interconnected systems is significant.The main contributions of the thesis include the following two aspects:1.The thesis proposes the optimal decentralized control scheme for nonlinear systems with mismatched interconnection of unknown time-delays.A local robust observer is employed to identify the partially unknown dynamics and release the assumption of matching interconnection.A novel local value function is constructed to eliminate the effects caused by the substitution error,the observation error and the mismatched interconnection.Based on robust-observer critic-structure,the local Hamilton-Jacobi-Bellman(HJB)equation is solved by adaptive critic design to obtain the optimal decentralized control laws.The stability of the closed-loop nonlinear system with mismatched time-delayed interconnection is proved by Lyapunov's direct method.Finally,two simulations are provided to verify the effectiveness of the optimal decentralized control scheme.2.The thesis develops an event-driven mechanism-based optimal decentralized tracking control scheme for unknown nonlinear interconnected system subject to input constrains.A non-quadratic utility function is introduced to eliminate the effect of input constrains.Then,a neural-network-based local observer is constructed to identify the subsystem dynamics.To deal with the substitution error,the observation error and the trajectory tracking error,a novel local value function is established.By implementing the local critic network to approximate the local value function,the optimal decentralized control law is obtained by solving the HJB equation.In order to save the computational resources and communication bandwidths,the aperiodic update mechanism is applied on the local observer and the local critic network.Furthermore,the local control law is executed and updated simultaneously when the event is triggered.Through Lyapunov stability theory,the event-driven decentralized tracking control scheme guarantees the stability of the closed-loop nonlinear interconnected systems with input constrains.Finally,two simulations are utilized to verify the effectiveness of this scheme.Finally,the conclusion and perspective of future research are provided at the end of this thesis.