Model-Free Adaptive Control Approach Using Integral Reinforcement Learning
Mohammed Abouheaf, Wail Gueaieb · 2019
Integral reinforcement learning control approaches with derivative weighting performance indices require full knowledge of dynamic models of the considered systems. These approaches do not provide straightforward solutions for underlying integral Bellman optimality equations. This urged for innovative online model-free processes with simple adaptation mechanisms. An online integral reinforcement learning control approach is developed herein for systems operating in uncertain dynamical environments. It employs a value iteration adaptation process to solve the underlying integral temporal difference equation accompanied by model-free optimal control strategies. The proposed approach is tested to control a flexible wing aircraft where the system dynamics are not required by the online learning process. The stability and convergence properties of the adaptive learning mechanism are formally proven before they are validated through numerical simulations.