Research on Intelligent Power Grid Attack Detection System Based on Machine Learning

The development of smart grids, coupled with the advancements in communication, network, and automation control technologies, has led to the integration of physical devices in power systems with communication networks, enhancing efficiency and intelligence. However, this integration also makes the power system more susceptible to network attacks. This study is dedicated to detecting False Data Injection Attacks (FDIA) in smart grids using machine learning. A large amount of node state variable sample data was employed to train Random Forest classifiers and XGBoost algorithms, with parameter optimization significantly enhancing their accuracy in various scenarios. The study found that while Random Forest excels in parallel training and noise resistance, XGBoost is more suited for short-term state prediction based on historical data. Both algorithms demonstrated effective performance and high accuracy in FDIA detection through extensive simulation. Furthermore, the paper explores the construction, implementation, and detection methods of FDIA, vital for the safe operation of smart grids. The research in this paper not only provides robust cybersecurity measures for smart grids but also lays the groundwork for future research on FDIA in more complex power system environments.