A Deep Generative Active Learning-based Scheme for the Efficient Development of Transient Stability Predictor

In this paper, we present a deep generative active learning (DGAL)-based scheme for transient stability prediction. Considering the high computational cost of time-domain simulation for transient stability assessment, a neural process is adopted to predict transient stability status, while a Monte Carlo feedforward is performed to assess the central tendency and dispersion of the predicted outcome. A combination of a generative model and an active learning tactic is employed to achieve data augmentation, while the samples with the highest degree of importance for annotation are selected, which not only greatly reduces the computational overhead associated with transient stability assessment but also improves prediction performance of the DGAL predictor. An examination of an 8169-node power system in a specific Chinese locale shows that the scheme can both improve training efficiency and meet prediction accuracy requirements.