Fast Generation of Power-Flow Feasible Solutions via Secretary Bird Optimization and Physics-Embedded Modeling
TL;DR
A Secretary Bird Optimization Algorithm-assisted physics-embedded learning method that achieves lower constraint violations, requires fewer labeled samples, and generalizes better to unseen operating conditions than benchmark methods.
Abstract
: Fast generation of physically feasible optimal power flow (OPF) solutions is essential for online power system operation under complex conditions. However, purely data-driven models may violate constraints, while existing physics-guided methods often rely on manually tuned loss weights, limiting their adaptability. This paper proposes a Secretary Bird Optimization Algorithm (SBOA)-assisted physics-embedded learning method. By embedding power-flow equality and operational inequality constraints into the neural-network training objective, the model learns the mapping from operating conditions to OPF solutions while enhancing physical feasibility. Moreover, an SBOA-based adaptive weight optimization strategy is introduced to coordinate supervised prediction, power-flow balance, and inequality-constraint losses, reducing empirical tuning and balancing optimality with feasibility. Results on multiple test systems show that the proposed method achieves lower constraint violations, requires fewer labeled samples, and generalizes better to unseen operating conditions than benchmark methods.