Knowledge Transferred DRL-Based Adversary for Cyberattacks on Active Distribution Network Volt-Var Control Agents:...

Area of Science:

• Electrical Engineering
• Cybersecurity
• Artificial Intelligence

Background:

• Deep reinforcement learning (DRL) based volt-var control (VVC) in active distribution networks (ADNs) is vulnerable to adversarial cyberattacks.
• Existing attacks often neglect the bad data detection (BDD) mechanism and assume unlimited attacker resources, limiting their real-world applicability.

Purpose of the Study:

• To develop a novel DRL-based adversary capable of launching strategically timed and stealthy attacks against VVC agents in ADNs.
• To maximize voltage violations in ADNs by optimizing attack timing and ensuring stealth against BDD mechanisms.

Main Methods:

• Reformulating the adversarial attack as a hybrid-action state-adversarial Markov decision process (MDP).
• Designing a new objective function to maximize voltage violations with minimal attacks, incorporating attack timing.
• Implementing perturbation set transformation on attack vectors to bypass BDD and ensure stealth.

Main Results:

• The proposed DRL adversary successfully generated strategically timed and stealthy adversarial samples.
• These samples significantly impacted the voltage regulation capability of VVC agents in ADNs.
• The attacks demonstrated effectiveness in bypassing the BDD mechanism, outperforming baseline methods.

Conclusions:

• The developed DRL-based adversary provides a robust method for evaluating the security of DRL-based VVC in ADNs.
• Strategic timing and stealth are crucial for effective adversarial attacks against VVC systems.
• This research highlights the need for enhanced BDD mechanisms resilient to sophisticated adversarial attacks.