Abstract
Transient stability assessment (TSA) is pivotal to power system security and stability, with online TSA providing core support for operators to make emergency control decisions. While deep learning has driven significant advancements in TSA, most existing deterministic models fail to quantify output uncertainty—a critical factor for evaluating prediction confidence and enabling operators to make informed decisions. This paper proposes a Bayesian Convolutional Neural Network (BCNN)-based TSA method that enables uncertainty estimation. By modeling the weights of network parameters as Gaussian distributions and adopting the Bayesian backpropagation algorithm for weight updates, the approach enhances TSA’s robustness against PMU noise and missing data. A case study on the New England 39-bus system demonstrates the model’s superior adaptability to small sample sizes and effective uncertainty estimation. Furthermore, the confidence intervals of evaluation results serve as valuable support for operators’ decision-making.