Research On Robust Control Of A Class Of Nonlinear Systems Based On Experience Replay

Optimal control has long been a significant research topic in the field of control.Due to the nonlinearity of most real-world systems,the study of optimal control for nonlinear systems has garnered increasing attention.As technology advances with the demands of daily production,engineering systems continue to expand in scale and complexity,making it increasingly challenging to obtain precise mathematical models.However,traditional control methods rely on accurate model information to design a valid control solution.Thus,there is a need to develop novel intelligent control methods for solving optimal control problems in the absence of model information.Adaptive dynamic programming(ADP)is an effective approach for addressing optimal control problems in the cases that no precise models can be constructed for control systems.ADP leverages reinforcement learning,dynamic programming,and neural network theories to construct an actor-critic architecture based on neural networks.This architecture iteratively approximates the cost function and control strategy in solving an optimal control problem.Meanwhile,it is desirable that the resulting controllers could possess the ability to attenuate the effects of disturbance.In this thesis,we propose an ADP method based on an experience replay to find an optimal and robust control strategy for continuous-time nonlinear systems taking into account disturbance.The proposed approach achieves closed-loop stability while maintaining desired performance.The main research contributions include:(1)An adaptive dynamic programming algorithm based on the actor-critic network structure with an experience replay is proposed for the robust control in nonlinear systems.The algorithm approximates the solution of the HJI equation by constructing a neural network,and then applies the adaptive dynamic programming method to iteratively update and find the optimal control strategy,so that the controlled system can achieve closed-loop stability while maintaining desired performance.Finally,the system stability analysis is given and the effectiveness of the algorithm is verified by simulation.(2)An adaptive dynamic programming algorithm based on the actor-critic network structure with an experience replay is proposed for the robust control of nonlinear systems with input constraints.The algorithm approximates the solution of the HJI equation by constructing a neural network,and then applies the adaptive dynamic programming method to iteratively update and find the optimal control strategy,so that the controlled system can achieve closed-loop stability while maintaining desired performance.Finally,the system stability analysis is given and the effectiveness of the algorithm is verified by simulation.