Reinforcement Learning Control For Uncertain Nonlinear Systems

Uncertain nonlinear system is a kind of difficult to precise control of the system,due to the complexity of system dynamic uncertainty,so that from the control theory of technology difficult to implement,and reinforcement learning has in the unknown environment interact with the environment of autonomous learning ability,can according to the system environment training model of optimal control strategy,It does not require prior knowledge of system environment information and is an ideal online learning algorithm for solving complex uncertain nonlinear system control.In order to solve the problem that a large number of uncertain nonlinear systems are difficult to be precisely controlled in real life,reinforcement learning control for a class of nonlinear systems is studied in this paper.This paper studied the reinforcement learning step control of the basic framework,the uncertainty of the system introduced reinforcement learning to perform network module,based on the Lyapunov function and the system performance index function for the system output tracking error and the adaptive error define returns status value function,system is necessary for the execution of network training efficiency is deduced error function;At the same time,the reinforcement learning evaluation network module is introduced to design the evaluation network model for the state value function in the system control,and the mean square residual function required by the evaluation network training is deduced by the Hamilton function.In the control process,the adaptive weight laws of the executive network and the evaluation network update the gradient through the effect error function and the objective principle of the minimum value of the mean square residual function.Then,in this paper,a class of nonlinear systems is the step control design ideas,through the analysis of the stability are derived based on Lyapunov step as well as the system’s control design,further design of each subsystem closed-loop system equation is deduced step inverse control law and a new law system with parameters,and combined with reinforcement learning of basic design framework,A reinforcement learning backstepping control method for nonlinear systems is proposed.Then,this paper studies reinforcement learning backstepping control for a class of uncertain nonlinear systems with matched or unmatched disturbances.The matching and unmatched disturbances of the system are compensated by using second-order HDOB and STDO disturbance observer.A reinforcement learning anti-disturbance backstepping control method for uncertain nonlinear systems is proposed.Finally,the reinforcement learning control for a class of Lipsitz nonlinear systems is studied.Based on the approximate Lipsitz theory and reinforcement learning backstepping control,a reinforcement learning backstepping control method for Lipsitz nonlinear systems is proposed.In order to verify the effectiveness and feasibility of the proposed control method,the intelligent control method based on reinforcement learning and backstepping control is simulated for multi-class nonlinear systems.The simulation results show that the adaptive parameters of the system are stable and convergent,and the proposed control method has the ability of adaptively fitting complex uncertain factors.Meanwhile,it is proved that the reinforcement learning control system is more effective and advantageous in realizing tracking error stability and invariance in uncertain adaptive environment.