Hybrid-Deep Ensemble Anomaly Intelligence for Secure and Resilient Power Grid

Abstract

Modern power networks are increasingly complex due to technological advancements, market deregulation, and ageing infrastructure, leading to heightened vulnerabilities and a critical need for robust anomaly detection mechanisms to ensure grid stability and resilience. Traditional anomaly detection methods and even machine learning techniques often fall short in adapting to the dynamic and intricate nature of these systems, struggling with high false alarm rates and limited detection accuracy, especially against sophisticated cyber threats. While deep learning models have emerged as promising solutions, existing approaches may not fully capture the complex feature space inherent in power system data for effective anomaly detection. To address these challenges, this project proposes a novel hybrid deep learning model, the Autoencoder Enhanced Random Forest, which leverages an Autoencoder for unsupervised feature extraction and reconstruction error calculation, subsequently used to augment the training data for a Random Forest classifier. Evaluated on the Power System Attack Dataset, a publicly available resource for cybersecurity research, and using accuracy, precision, recall, and F1-score as evaluation metrics, the proposed system achieved a high accuracy of 93%, precision of 92%, recall of 98%, and an F1-score of 95%. These results significantly outperform existing methods like Random Forest, AdaBoost, and JRipper, demonstrating the effectiveness of the Autoencoder Enhanced Random Forest approach for enhancing anomaly detection in modern power systems and contributing to improved grid security and operational reliability in the face of evolving threats.