Robust Training for AC-OPF (Student Abstract)

Electricity network operators use computationally demanding mathematical models to optimize AC power flow (AC-OPF). Recent work applies neural networks (NN) rather than optimization methods estimate locally optimal solutions. However, NN training data is costly and current cannot guarantee or feasible This study proposes a robust approach, which starts with small amount of seed uses iterative feedback generate additional in regions where the model makes poor predictions. The method applied non-linear univariate multivariate test functions, an IEEE 6-bus AC-OPF system. Results suggest can achieve prediction performance similar to, better than, regular training, while using significantly less data.