[Paper Review] Federated Learning for Smart Grid: A Survey on Applications and Potential Vulnerabilities

Zikai Zhang, Suman Rath+2 more|arXiv (Cornell University)|Sep 16, 2024

Privacy-Preserving Technologies in Data4 citations

TL;DR

This survey presents a comprehensive analysis of federated learning (FL) applications in smart grids across generation, transmission/distribution, and consumption stages, emphasizing privacy-preserving collaborative training. It identifies key vulnerabilities—such as Byzantine, backdoor, and inference attacks—and proposes robust aggregation, differential privacy, and anomaly detection as core defenses to bridge the gap between FL research and practical smart grid deployment.

ABSTRACT

The Smart Grid (SG) is a critical energy infrastructure that collects real-time electricity usage data to forecast future energy demands using information and communication technologies (ICT). Due to growing concerns about data security and privacy in SGs, federated learning (FL) has emerged as a promising training framework. FL offers a balance between privacy, efficiency, and accuracy in SGs by enabling collaborative model training without sharing private data from IoT devices. In this survey, we thoroughly review recent advancements in designing FL-based SG systems across three stages: generation, transmission and distribution, and consumption. Additionally, we explore potential vulnerabilities that may arise when implementing FL in these stages. Furthermore, we discuss the gap between state-of-the-art (SOTA) FL research and its practical applications in SGs, and we propose future research directions. Unlike traditional surveys addressing security issues in centralized machine learning methods for SG systems, this survey is the first to specifically examine the applications and security concerns unique to FL-based SG systems. We also introduce FedGridShield, an open-source framework featuring implementations of SOTA attack and defense methods. Our aim is to inspire further research into applications and improvements in the robustness of FL-based SG systems.

Motivation & Objective

• To systematically review federated learning applications in smart grids across generation, transmission/distribution, and consumption stages.
• To identify and analyze potential security vulnerabilities specific to FL-based smart grid systems, including Byzantine, backdoor, and inference attacks.
• To highlight the gap between state-of-the-art FL research and real-world deployment in smart grids, particularly concerning non-IID data and computational efficiency.
• To propose future research directions focused on robust aggregation, adaptive model optimization, and advanced defense mechanisms for FL in critical energy infrastructure.

Proposed method

• Categorizes FL applications in smart grids into three stages: generation (renewable forecasting), transmission/distribution (fault detection, optimal power flow), and consumption (load forecasting, energy theft detection).
• Analyzes vulnerabilities in FL-based SG systems, including model poisoning via Byzantine attacks, backdoor attacks, and inference attacks on model updates.
• Evaluates defense mechanisms such as robust aggregation (Krum, Bulyan, Trimmed-Mean), differential privacy (FedVRDP), and anomaly detection for malicious client identification.
• Proposes adaptive techniques to handle non-IID data distributions across IoT devices in SGs, improving model convergence and accuracy.
• Recommends client selection strategies based on data quality and computational capability to enhance training efficiency and model robustness.
• Integrates secure aggregation and privacy-preserving techniques like homomorphic encryption and secure multi-party computation to mitigate inference attacks.

Experimental results

Research questions

• RQ1 How can federated learning be effectively applied across the three stages of smart grid systems—generation, transmission/distribution, and consumption—while preserving data privacy?
• RQ2 What are the primary security vulnerabilities in FL-based smart grid systems, and how do they differ from those in centralized machine learning approaches?
• RQ3 How do non-IID data distributions across smart grid IoT devices impact the performance and convergence of FL models, and what techniques can mitigate this?
• RQ4 What defense strategies are most effective against Byzantine, backdoor, and inference attacks in FL-based smart grid deployments?
• RQ5 What key research gaps exist between current FL methodologies and their practical implementation in real-world smart grid environments?

Key findings

• Federated learning enables privacy-preserving collaborative model training in smart grids without sharing raw data, significantly enhancing data security in distributed energy systems.
• Non-IID data across smart grid IoT devices remains a major challenge, degrading FL model accuracy and requiring adaptive techniques for effective convergence.
• Byzantine attacks pose a critical threat by injecting malicious model updates that can corrupt the global model, necessitating robust aggregation methods like Krum and Bulyan.
• Backdoor attacks can be stealthily embedded into FL models, leading to system-wide failures if undetected, highlighting the need for enhanced anomaly detection and differential privacy.
• Inference attacks remain a concern even in FL, as adversaries may extract sensitive information from model updates, requiring stronger privacy-preserving mechanisms such as homomorphic encryption.
• Hybrid defense frameworks like FedVRDP, combining differential privacy and Byzantine resilience, show promise in providing dual-layered protection against both privacy and integrity threats.