Deep Learning-Based Optimal Power Flow Algorithm with Grid Topology-Variation Perception

With the increasing penetration of renewable energy resources and the continuous diversification of power system operating conditions, data-driven methods are becoming an important means for real-time optimal power flow (OPF) decision-making in power system planning and operation because of their capability to process large-scale data and provide fast inference. However, under stochastic operating conditions and network topology changes, existing methods still face several key challenges, including limited physical interpretability, insufficient training samples for newly encountered topologies, and inadequate generalization across varying network structures. To address these issues, this paper proposes a deep learning-based OPF method for power systems with topology changes. Multiple topology domains are constructed for source-domain model pretraining, while Chebyshev graph convolution weighted by branch-admittance magnitudes is employed to extract multi-order electrical coupling features under different network topologies. When an unseen topology is encountered, the parameters of the pretrained source-domain model are used to initialize the target-domain model, and the graph propagation operator is reconstructed according to the target topology. Physics-guided information is further incorporated into the training objective, and the model is fine-tuned using a limited number of target-domain samples to obtain a topology-adaptive model. Simulation studies on the IEEE 118-bus test system demonstrate the effectiveness and superior performance of the proposed method.