Decentralized intelligence in smart grids: a comprehensive survey of edge computing and federated learning
2666-5468
Abstract
Modern power grids are becoming more distributed and data-intensive, while centralized, cloud-based analytics struggle to keep pace. Routing everything through a central system introduces delays, strains network capacity, complicates compliance with data protection rules, and concentrates cybersecurity risk at a single point of failure. Edge computing and federated learning (FL) have emerged as complementary responses, moving computation and learning closer to where grid data is generated while keeping that data in place. This comprehensive survey synthesizes the architectural, algorithmic, and operational dimensions of Edge–FL integration in smart grids. It first situates distributed intelligence within cyber–physical grid architectures and heterogeneous communication fabrics. It then systematically examines FL paradigms, including horizontal and vertical federation, adaptive optimization under non-IID conditions, and privacy-preserving mechanisms, and analyzes their alignment with edge-constrained environments. Architectural integration models spanning hierarchical, peer-to-peer, and hybrid edge–cloud coordination are critically evaluated alongside communication-efficient aggregation, secure computation techniques, and resilience mechanisms. The survey further consolidates application domains, including load forecasting, distributed energy resources coordination, anomaly detection, predictive maintenance, and market participation, using multidimensional evaluation criteria that extend beyond accuracy to encompass latency, scalability, robustness, and deployment feasibility. Real-world pilot initiatives and industrial platforms are comparatively analyzed to assess implementation maturity, interoperability gaps, and regulatory readiness. Finally, the paper outlines open research directions in adversarial robustness, federated reinforcement learning, TinyML-enabled edge adaptation, sustainability-aware orchestration, and emerging cross-layer co-design frameworks. The review identifies the conditions under which Edge–FL can support decentralized grid intelligence and the barriers that currently limit utility-scale deployment.