Using Recurrent Neural Networks to improve initial conditions for a solar wind forecasting model

• DOI: 10.1016/j.engappai.2024.108266
• Publication Date: 2024-03-15T17:44:44Z
• Authors: Filipa S. Barros; Paula A. Graça; J.J.G. Lima; Rui F. Pinto; André Restivo; Murillo Villa
• Keywords: Robustness

Solar wind forecasting is a core component of Space Weather, field that has been the target many novel machine-learning approaches. The continuous monitoring Sun provided an ever-growing ensemble observations, facilitating development models predict solar properties on Earth and other celestial objects within system. This enables us to prepare for mitigate effects wind-related events space. performance some simulation-based depends heavily quality initial guesses used as conditions. work focuses improving accuracy these conditions by employing Recurrent Neural Network model. study's findings confirmed Networks can generate better simulations, resulting in faster more stable simulations. In our experiments, when we predicted conditions, simulations ran average 1.08 times faster, with statistically significant improvement reduced amplitude transients. These results suggest improved enhance numerical robustness model enable moderate integration time step. Despite modest simulation convergence time, model's reusability without retraining remains valuable. With lasting up 12 h, 8% gain equals one hour saved per simulation. Moreover, generated profiles closely match simulator's, making them suitable applications less demanding physical accuracy.