Learning solar flare forecasting model from magnetograms

Abstract

Solar flare is one type of violent eruptions from the Sun. Its effects almost immediately arrive to the near-Earth environment, so it is crucial to forecast solar flares in space weather. So far, the physical mechanisms of solar flares are not yet clear, hence we learn a solar flare forecasting model from the historical observational magnetograms by using the deep learning method. Instead of designing the feature extractor by the solar physicist in the traditional solar flare forecasting model, the proposed forecasting model can automatically learn features from input raw data, and followed by a classifier for foretasting from the learned features. The experimental results demonstrate that the proposed model can achieve better performance of solar flare forecasting comparing to traditional solar flare forecasting models.