Home // SECURWARE 2024, The Eighteenth International Conference on Emerging Security Information, Systems and Technologies // View article
Authors:
Antonin Delhomme
Livinus Obiora Nweke
Sule Yildirim Yayilgan
Keywords: Smart Grids; Digital Substation; Machine Learning; Deep Learning; DoS Attacks; Cyber-Attack Detection.
Abstract:
The increasing digitalization of power grids, often referred to as smart grids, has revolutionized the efficiency and functionality of electrical infrastructure. Smart grids integrate advanced communication technologies and digital controls to optimize the generation, distribution, and consumption of electricity. However, this digital transformation has also introduced significant cybersecurity challenges. As these grids are critical national infrastructures, ensuring their protection against cyber threats is essential. This study investigates the application of various machine learning algorithms to detect Denial of Service (DoS) attacks within the IEC 61850 communication protocols, specifically Generic Object-Oriented Substation Event (GOOSE) and Sampled Values (SV). We employed a simulated substation communication environment to generate normal and attack scenarios, utilizing both GOOSE and SV messages. The machine learning models used in our experiment include a Random Forest Classifier, Decision Tree, Support Vector Machine (SVM), Neural Networks, K-Nearest Neighbors (KNN), Logistic Regression, Gradient Boosting, and a Voting Classifier. The results demonstrated that the Random Forest Classifier and Decision Tree models consistently achieved high accuracy and F1 scores, making them effective for DoS detection in IEC 61850 protocols. The Voting Classifier also showed strong performance, leveraging the strengths of multiple models. Despite the generally good performance of these models, the SVM and Voting Classifier provided the best results in a specific instance with reduced data volume. Training time was also considered, highlighting Decision Tree and Logistic Regression as the most efficient models for quick deployment. This study underscores the potential of machine learning-based approaches for enhancing the security of substation communication systems, providing valuable insights for future research and practical applications in the field of smart grid cybersecurity.
Pages: 1 to 8
Copyright: Copyright (c) IARIA, 2024
Publication date: November 3, 2024
Published in: conference
ISSN: 2162-2116
ISBN: 978-1-68558-206-7
Location: Nice, France
Dates: from November 3, 2024 to November 7, 2024