Self-Adaptive Federated Meta-Learning Framework for Hybrid Renewable Microgrid Orchestration under High Renewable Penetration

Abstract

The rapid expansion of renewable energy resources has fundamentally transformed the operational dynamics of distributed power systems. Hybrid microgrids integrating photo- voltaic arrays, wind turbines, and battery storage offer localized resilience and improved energy efficiency. However, renewable intermittency introduces significant uncertainty in forecasting and dispatch planning, particularly under high penetration sce- narios. Conventional energy management systems rely on static weighting strategies and centralized machine learning models, limiting adaptability and scalability in heterogeneous microgrid environments.

This paper proposes a unified self-adaptive renewable energy orchestration framework that integrates meta-learning, federated gradient aggregation, and convex hybrid dispatch optimization within a hierarchical edge–fog–cloud architecture. An adaptive renewable weighting mechanism is mathematically formulated using softmax-based meta-parameterization, enabling dynamic regulation of solar, wind, and storage contributions. A distributed convex optimization problem is constructed for each microgrid, ensuring bounded state-of-charge dynamics under operational constraints. Federated learning enables cross-microgrid parame- ter adaptation without raw data exchange, preserving scalability and data locality.

Formal stability guarantees are established under bounded learning rates and convex feasibility conditions. Semi-realistic dataset-driven simulations involving multiple hybrid microgrids demonstrate statistically significant improvements in forecasting accuracy, renewable utilization, and operational cost compared to static and centralized baselines. The proposed framework provides a scalable and resilient foundation for intelligent de- centralized renewable energy management systems.