A Novel Data-Driven Transient Stability Assessment of Power Systems Based on Uncertain Data Generation

Transient stability analysis is crucial in power systems, and thus, data-driven methods for transient stability analysis have received extensive research attention due to their accuracy and efficiency in existing studies. However, traditional approaches primarily focus on improving models to enhance the accuracy of transient stability assessment, often overlooking the significance of the dataset itself. In response to this gap, this paper introduces a data-centric perspective to augment power system transient stability analysis. By employing adversarial training, our proposed method generates uncertain samples based on the given transient stability analysis model and then incorporates these samples into the training dataset, which could significantly improve the accuracy and robustness of the existing data-driven models. The proposed method was tested on the New England 39-bus system to demonstrate its effectiveness. The results indicate that enriching the training dataset with generated samples to challenge and refine the model provides a new perspective for improving data-driven transient stability analysis.