Resiliency Assessment of Power Systems Using Deep Reinforcement Learning

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Power system resilience is critical for reliable operation during abnormal conditions.
• Assessing resilience aids in developing effective planning and operational strategies.
• Sequential topology attacks pose a significant threat to power system stability.

Purpose of the Study:

• To introduce a novel measure, the level-of-resilience (LoR), for quantifying power system resilience.
• To evaluate the effectiveness of various deep reinforcement learning (DRL) agents in determining LoR.
• To identify the most efficient DRL agent for power system resiliency assessment.

Main Methods:

• Development and application of a level-of-resilience (LoR) metric.
• Utilizing four deep reinforcement learning (DRL) agents: deep Q-network (DQN), double DQN, REINFORCE, and REINFORCE with baseline.
• Conducting case studies on the IEEE 6-bus test system to simulate sequential topology attacks.

Main Results:

• The double DQN agent demonstrated the highest success rate in determining the LoR.
• The double DQN agent exhibited the fastest performance compared to other DRL agents.
• The proposed LoR measure effectively quantifies power system vulnerability under attack scenarios.

Conclusions:

• Deep reinforcement learning, particularly the double DQN agent, is a highly efficient tool for power system resiliency evaluation.
• The LoR measure provides a valuable metric for understanding and enhancing power system robustness.
• Findings support the adoption of advanced AI techniques for critical infrastructure protection.