Deep Active Learning-driven Acceleration of Smart Grids Security Assessment

Reliable and efficient operation of smart grids (SGs) is contingent on fast and accurate security assessment, which is increasingly difficult because of the growing complexity and uncertainty of modern power grids. As traditional power flow (PF) simulations are computationally intensive, machine learning (ML)-based approaches have gained traction. However, such models typically require large amounts of labelled data, which is time-consuming to acquire via PF oracles. To mitigate this, we propose a deep active learning (DAL)-driven framework that actively selects the most informative operational points (OPs) for labelling, thereby reducing reliance on exhaustive PF computations. We evaluate multiple DAL query strategies—including Monte Carlo (MC) dropout and batch active learning by diverse gradient embeddings (BADGE)—on a binary classification task to detect congestion in the IEEE European low voltage test network (ELVTN). Results show that DAL methods significantly reduce the number of training labelled samples required over the random baseline on the considered SG security assessment dataset. Our findings suggest that DAL is a promising avenue for accelerating the training of ML-models in SGs by reducing dependence on costly PF-based labelling. The code used to reproduce the results presented in this paper is publicly available at https://gitlab.com/satie.sete/dal accel sg pf.