Optimal power flow under varying topologies via physics-guided neural network with stack-learning

As a critical tool in power system analysis, the rapid solution of AC Optimal Power Flow (AC-OPF) has consistently garnered significant attention. Traditional algorithms require substantial time to calculate OPF problems, while existing neural network approaches, though faster, are often limited to fixed test system, thus not suitable for varying topologies. To overcome this difficulty, this paper proposes a Physics-Guided Stacked Attention and Convolutional Neural Network (PG-SAC) method to get accurate and fast predictions in AC-OPF problems with N-1 contingency. The PG-SAC method integrates physical principles to comprehensively consider various physical quantities within the power system. Besides, Through the combination of convolutional neural networks (CNNs) and self-attention mechanisms, the model captures both local and global features, followed by a stacking technique to refine the OPF solution. Extensive tests on different IEEE bus systems show that the proposed method outperforms the current state-of-the-art, achieving a 76.05% reduction in objective-function RMSPE and an average 81-fold acceleration in solution speed compared to traditional techniques across all test scenarios. These results validate the efficacy of the method and highlight its substantial potential in addressing topology-variable AC-OPF challenges. • A physics-guided deep learning method is proposed for solving OPF under varying topologies. • The PG-SAC model embeds physical laws and improves accuracy via stack-learning refinement. • The approach ensures scalability and generalization in large-scale power systems.