Feasible Operating Region-Based Sampling for Representative OPF Datasets

The increasing interest in data-driven approaches for solving the AC Optimal Power Flow (ACOPF) problem has highlighted the importance of representative datasets. These datasets are crucial for benchmarking the surrogate models, particularly under realistic and previously unobserved grid conditions. However, generating such datasets remains a significant challenge due to the non-convexity of the ACOPF problem, as well as limitations related to data availability, privacy, and confidentiality. In this work, we present a novel constraint-guided exploration of the Feasible Operating Region (FOR), aimed at efficiently identifying and sampling regions where system constraints become active. To generate corresponding load profiles, we employ an OPF-based disaggregation method that ensures physical and operational feasibility. We validate the proposed approach on SimBench benchmark grids, demonstrating its effectiveness in capturing active constraint regions when compared to random sampling methods.