FQL_Puma: Integrating fuzzy reinforcement learning approach into PUMA optimization for adaptive search strategy and toward energy aware edge computing

Meta-heuristic algorithms need to balance exploration and exploitation to find good solutions. Most algorithms use fixed rules for this balance, which limits their performance. This paper presents FQL_Puma, an enhanced version of the Puma Optimizer that integrates fuzzy logic with Q-learning to adaptively select search strategies during the optimization process. To clearly assess the contribution of the fuzzy mechanism, a non-fuzzy Q-learning variant (QL_Puma) is also developed for comparison. This paper, also, evaluates both QL and FQL variants for the DIO algorithm. In addition to benchmark optimization, this study addresses sustainability in edge computing, where the rapid growth of connected devices and data processing has led to increased energy consumption. A system model is developed that jointly considers the energy consumption of both edge devices and edge servers in task offloading and resource allocation. The proposed methods are evaluated on standard benchmark functions and a realistic edge computing scenario. Experimental results show that the FQL-based strategy selection improves the original PUMA. Specifically, FQL_Puma improves PUMA’s performance on 14 to 18 out of 29 benchmark functions across different dimensions and achieves lower total energy consumption than PUMA in almost all evaluated edge computing scenarios.