Dynamic Adaptive Fuzzy Modeling (DAFM) is a flexible framework for describing nonlinear, time-varying systems, yet its parameters must adapt reliably to changing conditions. We propose a reinforcement-learning (RL) approach that performs system identification online by embedding a temporal-difference learning rule with eligibility traces into the DAFM update law. The method updates membership and consequent parameters directly from streaming data, assigning greater credit to recent errors while remaining interpretable through the fuzzy rule base. We validate the algorithm on two nonlinear numerical examples and a laboratory liquid-level rig. Across these studies, the RL-driven DAFM achieves real-time identification with accurate tracking and strong robustness to nonstationary dynamics, demonstrating a practical route to data-efficient, interpretable, and adaptive fuzzy modeling.
Soliman et al. (Thu,) studied this question.