Answers are grounded in this paper and cite their sources.
Power system security assessment is crucial for ensuring grid stability and reliability. Recent machine learning approaches have shown promise for accelerating this task, but their vulnerability to measurement noise and communication disturbances limits practical deployment. This paper presents a unified robust training framework that addresses this challenge through multi-noise augmentation, boundary-aware sample selection, noise-invariant feature learning, and adaptive noise scheduling. We evaluate our approach on a provincial-level power system with 132 buses under various noise types including Gaussian, impulse, and adversarial perturbations. Results demonstrate that our framework achieves 96.1% accuracy on clean data while maintaining 91.4% average accuracy under noisy conditions—significantly outperforming standard training and existing robust methods. Notably, our approach reduces false negative rates from 28.4% to 9.5% under adversarial perturbations, addressing a critical concern for security assessment. The proposed framework enables more reliable application of machine learning for power system operations under real-world measurement uncertainties.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: