A Transformer-Based Framework for Geomagnetic Activity Prediction

Geomagnetic activities have a crucial impact on Earth, which can affect spacecraft and electrical power grids. Geospace scientists use geomagnetic index, called the Kp to describe overall level of activity. This index is an important indicator disturbances in Earth’s magnetic field used by U.S. Space Weather Prediction Center as alert warning service for users who may be affected disturbances. Early accurate prediction essential preparedness disaster risk management. In this paper, we present novel deep learning method, named KpNet, perform short-term, 1–9 hour ahead, forecasting based solar wind parameters taken from NASA Science Data Coordinated Archive. KpNet combines transformer encoder blocks with Bayesian inference, capable quantifying both aleatoric uncertainty (data uncertainty) epistemic (model when making predictions. Experimental results show that outperforms closely related machine methods terms root mean square error R-squared score. Furthermore, provide data model quantification results, existing cannot offer. To our knowledge, first time transformers been prediction.