A hybrid evolutionary learning framework for resource optimization in federated learning over heterogeneous 6G IoT networks

The emergence of sixth-generation (6G) networks and the rapid proliferation of Internet of Things (IoT) devices have introduced significant challenges for federated learning (FL), particularly in achieving energy-efficient resource allocation across heterogeneous wireless environments. To address these challenges, this study proposes the Hybrid Evolutionary Learning Framework (HELF), a novel, adaptive, and scalable approach for optimizing resource management in federated multitask learning (FMTL) systems. HELF integrates three established evolutionary optimization techniques, Adaptive Genetic Algorithms (AGA), Enhanced Particle Swarm Optimization (EPSO), and Differential Evolution (DE) to dynamically allocate subcarriers, regulate transmission power, and distribute computational workloads across a hierarchical network architecture comprising macro and micro base stations. By incorporating real-time adaptation mechanisms and population diversity control, HELF effectively addresses network heterogeneity and dynamic channel conditions, outperforming conventional static and heuristic resource allocation strategies. Extensive simulations conducted under realistic 6G IoT scenarios demonstrate that HELF reduces energy consumption and communication latency by approximately 10–12% compared with existing benchmark resource allocation approaches while maintaining or improving the convergence performance of federated learning models. Furthermore, the proposed framework exhibits strong scalability, supporting large-scale deployments involving diverse device capabilities and network conditions. The results highlight the potential of HELF to enhance the efficiency, responsiveness, and sustainability of next-generation wireless federated learning systems operating within complex IoT ecosystems.