Neuro-fuzzy control using reinforcement learning

This paper proposes a general control strategy that combines reinforcement learning with approximate reasoning-based methods. We use a neuro-fuzzy controller, because of its ability to capture human knowledge in the form of fuzzy IF-THEN rules. Starting from a roughly tuned set of rules, we propose an on-line self-tuning method, using only a simple real signal to evaluate the current process state and to tune the controller parameters. This method is applied to an unstable second order system and demonstrates good performances.<
>