Proactive scheduling of resilient networked microgrids via data-efficient security region learning

Networked microgrids (NMGs) have emerged as a key paradigm for enhancing the flexibility of distribution systems with high penetration of renewable energy. However, the rising frequency of natural disasters and physical attacks poses urgent challenges for improving NMGs resilience. Conventional proactive scheduling methods often focus solely on economic objectives while overlooking the multi-dimensional security requirements among microgrids. To address these challenges, this paper proposes a security region-oriented rolling proactive scheduling framework for resilient NMGs. The proposed approach first introduces a data-driven security region characterization algorithm based on the maximum-margin principle, which combines dual-space decomposition with entropy-driven active learning to rapidly and accurately identify security boundaries at minimal sampling cost. On this basis, a rolling two-stage stochastic scheduling model is developed, in which the security margin is adaptively incorporated and weighted according to transient frequency simulations of islanded microgrids. Case studies on modified IEEE 39-bus and 118-bus systems demonstrate that the proposed method significantly enhances scheduling robustness and operational resilience while reducing both computational and sampling burdens, thereby providing a practical and efficient solution for resilient NMG operation under evolving disaster scenarios. • A novel data-driven security region characterization algorithm for NMGs is proposed, which defines the security boundary via the maximum-margin principle. • An entropy-driven active learning algorithm is developed to enable rapid and accurate identification of security boundaries, significantly reducing the sample complexity and simulation burden under complex operational constraints. • A security region-oriented rolling proactive scheduling framework is established, in which the weighting of the security margin is dynamically adjusted based on frequency simulations of islanded microgrids.