Transferability-Oriented Adversarial Robust Security-Constrained Optimal Power Flow

Security-constrained optimal power flow (SCOPF) aims to achieve an economical operation while considering the security issues during contingencies. Data-driven security assessment methods can provide security constraints for SCOPF and have been widely developed in the smart grid community to deal with the challenges arising from the increased penetration of renewable energy resources and the deployment of power electronic devices. Nevertheless, it has been recognized that the machine learning model is vulnerable to adversarial examples, and the power system communication network is prone to cyber attacks, which affects the decision-making of security assessment and SCOPF. To this end, a transferability-oriented adversarial robust security-constrained optimal power flow (TOAR-SCOPF) is proposed to diminish the potential security risks from attacks, formulated as an attacker-defender problem. To alleviate the conservatism of TOAR-SCOPF, a more realistic scenario with unknown features and unknown selected machine learning models is considered in this work by developing a novel transferability-oriented adversarial attack (TOA) method for data-driven security assessment while considering the power system's physical constraints to bypass the bad data detection mechanism. Case studies are conducted based on the IEEE 39-bus and IEEE 68-bus systems, respectively, to explore the vulnerability and demonstrate the effectiveness of the proposed TOAR-SCOPF.