Federated Heuristic Co-Evolution: Adaptive WOA–ABC Fusion for Autonomous Edge Offloading

Abstract

Mobile Edge Computing (MEC) is becoming an essential platform for latency-critical Internet of Things (IoT), industrial, vehicular, and healthcare applications, which operate in heterogeneous environments with diverse, time-varying workloads, energy constraints, and communication regimes. Recent advances in task-offloading research have seen the rise of deep reinforcement learning, federated reinforcement learning, game-theoretic optimization, and hybrid metaheuristics; however, most works optimize either learning or offloading as a one-way process. This paper revisits and enhances the Federated Heuristic Learning (FHL) framework as a closed-loop co-evolutionary system that couples federated context learning and adaptive Whale Optimization Algorithm (WOA) and Artificial Bee Colony (ABC) search. A volatility index is used to drive a fuzzy controller that balances the exploration/exploitation ratio, while a volatility-aware privacy mechanism regulates update perturbation. The framework is formulated as a multi-objective task-offloading problem with minimum latency and energy as the optimization goals subject to resource, deadline, and assignment constraints. A comprehensive benchmark suite that includes five state-of-the-art 2022–2026 methods (federated, ensemble, and multi-agent deep reinforcement learning, adaptive multi-objective deep reinforcement learning, and joint time-energy federated offloading) is implemented, tuned, and evaluated under the same experimental settings. The numerical results demonstrate that FHL achieves lower energy consumption and latency, graceful scalability, and faster feedback adaptation than the five contemporary baselines, thus validating the closed-loop co-evolutionary approach.