A Comprehensive Survey on the Usage of Machine Learning to Detect False Data Injection Attacks in Smart Grids

Abstract

This article provides a comprehensive survey on the application of machine learning techniques for detecting False Data Injection Attacks (FDIA) in smart grids. It introduces a novel taxonomy categorizing detection methods based on key criteria such as AC and DC systems, performance metrics, bus size, algorithm selection, and specific subcategories of detection problems. The proposed taxonomy highlights the utility of Graph Neural Networks, autoencoders, and federated learning in addressing sub-problems like privacy preservation, generalized detection, locational detection, and attack classification. The survey underscores the importance of realistic, publicly accessible datasets and enhanced attack simulation techniques. Future research directions are suggested to further the development of robust FDIA detection methods in smart grids.