Data-Based Reinforcement Learning Approximate Optimal Control for an Uncertain Nonlinear System with Partial Loss of Control Effectiveness

Abstract

An infinite horizon approximate optimal control problem is developed for a system with unknown drift parameters and control effectiveness faults. A data-based filtered parameter estimator with a novel dynamic gain structure is developed to simultaneously estimate the unknown drift dynamics and control effectiveness fault. A local state-following approximate dynamic programming method with a modified update law and Bellman error extrapolation is used to approximate the unknown optimal value function for the uncertain system. A Lyapunov-based stability analysis shows uniformly ultimately bounded convergence for the closed-loop system.