Data-Driven Identification of Extreme Operating Scenarios in Power Systems

High renewable energy penetration makes power system operating scenarios increasingly complex and variable. Extreme operating scenarios have a low probability of occurrence and insufficient valid samples, which can cause class imbalance in the training data and weaken the risk identification capability of data-driven models. To address this problem, this paper proposes a method for extreme operating scenario generation, augmentation, and identification combined with external physical feasibility verification. An auxiliary classifier generative adversarial network is used to generate minority-class extreme scenarios in a targeted manner. Physical feasibility and risk-label consistency checks are then used to construct valid augmentation samples, and a support vector machine is trained to identify extreme operating scenarios. Case studies on the IEEE 118-bus and IEEE 300-bus systems show that the extreme-scenario recall of the proposed method exceeds 91% in both systems and outperforms direct Monte Carlo augmentation under the same real-sample and training-size budgets. The results indicate that the proposed method effectively alleviates extreme-scenario sample scarcity and class imbalance and improves extreme operating scenario identification.