Transformer-based Deep Learning Framework for Intrusion Detection in Smart Grids

Le résumé fourni par la source

As smart grids grow in complexity and interconnectedness, they become susceptible to more cyber threats that can impact grid operations and have great financial implications. Traditional IDS systems have a hard time detecting new-generation cyber-attacks, especially the ones that are sophisticated or novel. We present a new Transformer-based deep learning model for real-time smart grid intrusion detection in this paper. Our model uses the Transformer for anomaly detection and an Autoencoder for unsupervised learning to identify deviations from regular grid performance. The performance of the model is tested on the CIC IDS Dataset, and it is shown that our approach leads to much better detection precision as compared to existing state-of-the-art methods. In particular, the proposed model provides 95% detection accuracy, improving 20% over traditional signature-based IDS, with a false positive rate improvement of 15%. Moreover, the system can detect both short-lived intrusions (over 10s of seconds) like burst power spikes and long-term attacks (e.g., data exfiltration with a much slower creep), rendering it practically flexible for dynamic and sophisticated cyber hazards. The solution is scalable, can work as part of smart grid infrastructure, and presents a resource-efficient, real-time, comprehensive response against cyberattacks.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.