Scholarlink

<p>Smart grids increasingly depend on data-driven intelligence to enable real-time monitoring, adaptive control, and resilient energy management across large-scale and geographically distributed infrastructures. Although centralized machine learning approaches have demonstrated strong analytical capabilities, their practical deployment in smart grids is constrained by privacy concerns, regulatory and data ownership requirements, communication overhead, limited scalability, and vulnerability to single points of failure. Federated learning (FL) has emerged as a decentralized intelligence paradigm that enables collaborative model training while preserving data locality, making it particularly suitable for smart grid environments. Despite growing interest, existing studies on FL-based smart grid intelligence remain fragmented, with limited synthesis of architectural designs, operational integration, and deployment trade-offs. This review addresses this gap by systematically analyzing federated learning in smart grids from architectural, operational, and intelligence evolution perspectives. The progression from centralized and semi-decentralized learning toward hierarchical and fully federated intelligence is examined, highlighting how FL aligns with the physical and operational layers of smart grids, spanning consumer, distribution, transmission, and utility control levels. A unified taxonomy is introduced to characterize the evolution of federated intelligence, ranging from static aggregation schemes to adaptive, context-aware, self-evolving, and digital twin-driven frameworks. The review further consolidates experimental practices reported in the literature, including datasets, simulation platforms, evaluation metrics, and validation protocols, with particular emphasis on non-IID data handling, communication efficiency, privacy preservation, and robustness. Synthesized findings indicate that federated learning can significantly enhance privacy, scalability, and operational resilience when designed in accordance with grid topology, threat models, and real-time constraints. Finally, open challenges related to communication overhead, system heterogeneity, security, and deployment complexity are discussed, and future research directions toward adaptive and autonomous federated intelligence for next-generation smart grids are outlined.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The shift from centralized analytics to federated learning in smart grids: a comprehensive review

  • 期刊论文

  • 发表日期：

    2026年7月1日

查看全文
  • Arifa Sultana,
  • Kandarpa Kumar Sarma

摘要

Smart grids increasingly depend on data-driven intelligence to enable real-time monitoring, adaptive control, and resilient energy management across large-scale and geographically distributed infrastructures. Although centralized machine learning approaches have demonstrated strong analytical capabilities, their practical deployment in smart grids is constrained by privacy concerns, regulatory and data ownership requirements, communication overhead, limited scalability, and vulnerability to single points of failure. Federated learning (FL) has emerged as a decentralized intelligence paradigm that enables collaborative model training while preserving data locality, making it particularly suitable for smart grid environments. Despite growing interest, existing studies on FL-based smart grid intelligence remain fragmented, with limited synthesis of architectural designs, operational integration, and deployment trade-offs. This review addresses this gap by systematically analyzing federated learning in smart grids from architectural, operational, and intelligence evolution perspectives. The progression from centralized and semi-decentralized learning toward hierarchical and fully federated intelligence is examined, highlighting how FL aligns with the physical and operational layers of smart grids, spanning consumer, distribution, transmission, and utility control levels. A unified taxonomy is introduced to characterize the evolution of federated intelligence, ranging from static aggregation schemes to adaptive, context-aware, self-evolving, and digital twin-driven frameworks. The review further consolidates experimental practices reported in the literature, including datasets, simulation platforms, evaluation metrics, and validation protocols, with particular emphasis on non-IID data handling, communication efficiency, privacy preservation, and robustness. Synthesized findings indicate that federated learning can significantly enhance privacy, scalability, and operational resilience when designed in accordance with grid topology, threat models, and real-time constraints. Finally, open challenges related to communication overhead, system heterogeneity, security, and deployment complexity are discussed, and future research directions toward adaptive and autonomous federated intelligence for next-generation smart grids are outlined.