Abstract:

Smart grids, being advanced power grids that incorporate information and communication technologies, are under growing threats from sophisticated cyber attacks. Having strong intrusion detection capabilities in such grids is critical for the stability of operations and integrity of the system. Yet, conventional intrusion detection systems (IDS) tend to have poor generalizability to novel attack modes, limited adaptability, and high false positives, which makes them unsuitable for the evolving cyber-physical environment of smart grids. To overcome these deficiencies, this paper introduces a new framework called Resilient Deep Reinforcement Intrusion Detection System (R-DRIDS). R-DRIDS combines deep reinforcement learning (DRL) with cybersecurity resilience techniques to support improved detection features. The presented R-DRIDS framework is implemented to detect and respond against different cyber-physical attacks such as data injection, denial-of-service, and unauthorized access, in simulated smart grid scenarios. It utilizes real-time feedback and environmental interactions to enhance its detection rate and decrease response time across iterations. Experimental tests verify that R-DRIDS performs better than traditional IDS methods based on detection rate, adaptability, and robustness. It produces lower false positive results and increased accuracy, making it a promising solution for ensuring future smart grids are secure against emerging cyber attacks.