Data-physics joint-driven N-k static security assessment with higher-order graph convolution model

In power systems with a high penetration of renewable energy, data-driven N-k static security assessment ( N-k SSA) plays a critical role in contingency risk analysis. To improve prediction robustness and enhance the model’s capability to capture long-chain power flow transfers after disturbances, this paper proposes a data-physics joint-driven N-k SSA scheme ( N-k DPSSA). Specifically, a multi-hop graph convolution network (MPGCN) is introduced to expand the model’s receptive field, enabling better identification of long-chain power redistribution patterns. A branch feature extractor (BFE) and a branch power predictor (BPP) are designed to extract branch-level features and estimate branch power flows directly. Furthermore, a physics-informed Kirchhoff discriminator (KD) module is developed to verify prediction validity under relaxed Kirchhoff’s current law constraints and identify uncertain samples, thus improving overall robustness. The proposed DPSSA framework is validated on the IEEE 39-bus system, the IEEE 300-bus system, and a real provincial-level power system in China. Experimental results show that compared with traditional methods, the proposed approach achieves more than 2% improvement in accuracy on the test set and over 10% improvement in previously unseen topologies. Visualization of prediction outcomes further demonstrates the model’s superiority in handling long-chain power flow transfers and its resilience to structural variations, validating its potential for practical deployment in large-scale system security assessment. • A hybrid data-physics scheme enables robust N-k security assessment after local disturbances. • Multi-hop graph convolution captures long-chain power flow transfer patterns efficiently. • Branch-level feature embedding enables accurate power flow prediction. • Kirchhoff-constrained filtering improves prediction reliability under diverse scenarios.