Power System Transient Security Assessment using Unsupervised Probabilistic Deep Bayesian Neural Network

This paper introduces an unsupervised deep Bayesian network, built upon normalizing flow and deep Bayesian network principles, to precisely evaluate the transient security condition of a power system. The proposed approach can capture locational and temporal features using an imbalanced dataset, is noise-model-free, and can handle unlabeled data. It can learn interdependencies between different signals and understand high-dimensional signals in power systems. To validate its effectiveness, the proposed method is studied using the New England power system and shows accuracy and reliability in comparison with state-of-the-art deep networks (convolutional neural network (CNN) and long short-term memory (LSTM)) and shallow networks (support vector machine (SVM) and artificial neural network (ANN)).