Grid Transient Simulation Using Attention-Based Data Augmentation Technique with Supercomputing

Abstract

The stability of power systems, central to the unimpeded flow of daily life and economic activities in our modern world, is a critical aspect requiring precise forecasting. Notwithstanding, predicting such stability becomes an arduous task, especially amidst situations fraught with high complexity. To mitigate this, our study presents an avant-garde approach for transient simulation of power systems, incorporating Transformer-based data augmentation techniques. We proceed to delineate the application of Transformer models for data augmentation in our methodology. The ensuing augmented data is then used for training models to predict both the stability result and stability index of power systems. Comparative analysis between predictions sourced from original and augmented data indicates that the utilisation of Transformer data augmentation significantly boosts the accuracy of our forecasts. Additionally, we undertake an exhaustive examination of the prediction outcomes, enabling the identification of key factors that impact the stability of power systems. This paper, therefore, offers a groundbreaking and highly effective predictive method for power system stability, yielding a significant advancement in our understanding of power system dynamics and offering preemptive measures to counter potential instability.