Federated Learning for Privacy-Preserving Defense in Power Cyber-Physical Systems: Frameworks, Techniques, and Challenges

The increasing interconnection and digitalization of modern energy systems have intensified cybersecurity vulnerabilities in Power Cyber-Physical Systems (Power CPS). Traditional centralized defense approaches struggle to balance privacy preservation, scalability, and collaborative responsiveness across distributed infrastructures. Federated Learning (FL) emerges as a promising paradigm that enables distributed, privacy-preserving model training without sharing raw data. This review presents a comprehensive analysis of FL-based collaborative defense for Power CPS, spanning threat modeling, architectural taxonomies, privacy-preserving mechanisms, and real-world applications. We categorize FL techniques by learning structure, synchronization, and personalization, and examine privacy-enhancing technologies such as differential privacy, secure multiparty computation, homomorphic encryption, and trusted execution environments. Practical applications across substations, SCADA systems, WAMS, and EV infrastructures are reviewed alongside deployment challenges such as communication overhead, adversarial threats, and operational constraints. A roadmap is proposed for future research in cross-layer FL architectures, federated reinforcement learning, and regulatory standardization. The review concludes by advocating for cross-sector collaboration to operationalize federated defense as a cornerstone of resilient, secure, and privacy-compliant smart grids.