Robust preventive and corrective security-constrained OPF for worst contingencies with the adoption of VPP: A safe reinforcement learning approach

The rising frequency of extreme weather events calls for urgent measures to improve the resilience and reliability of power systems. This paper, therefore, presents a robust preventive-corrective security-constrained optimal power flow (PCSCOPF) model designed to strengthen power system reliability during N-k outages. The model integrates fast-response virtual power plants (VPPs), dynamically adjusting their injections to mitigate post-contingency overloads and maintain branch flows within emergency limits. Additionally, a novel approach combining deep reinforcement learning (DRL) with Lagrangian relaxation is introduced to efficiently solve the PCSCOPF decision-making problem. By framing the problem as a constrained Markov decision process (CMDP), the proposed Lagrangian-based soft actor-critic (L-SAC) algorithm optimizes control actions while ensuring constraint satisfaction during the exploration process. Extensive investigations have been conducted on the IEEE 30-bus and 118-bus systems to evaluate their computational efficiency and reliability.