Title: Robust Data-Driven Linear Power Flow Model with Probability Constrained Worst-Case Errors

Abstract: To limit the probability of unacceptable worst-case linearization errors that might yield risks for power system operations, this letter proposes a robust data-driven linear power flow (RD-LPF) model. It is applicable to both transmission and distribution systems and can achieve better robustness than the recent data-driven models. The key idea is to probabilistically constrain the worst-case errors through distributionally robust chance-constrained programming. It also allows guaranteeing the linearization accuracy for a chosen operating point. Comparison results with three recent LPF models demonstrate that the worst-case error of the RD-LPF model is significantly reduced over 2- to 70-fold while reducing the average error. A compromise between computational efficiency and accuracy can be achieved through different ambiguity sets and conversion methods.