Solar Flare Prediction with Recurrent Neural Networks

Abstract

As the star closest to Earth, the Sun offers a wealth of information on its own composition and behavior, as well as a basis for the composition and behavior of all stars. The Sun’s violent magnetism gives rise to various solar activity, including solar flares. A form of space weather, strong solar flares can damage communications and expose astronauts to dangerous radiation. Monitored nearly 24 hours a day by various satellites, the Sun’s magnetic field properties, observed over a period of time, are theorized to be indicative of an upcoming flare. A popular choice when working with time series data or sequences, recurrent neural networks (RNNs) are excellent for solar flare forecasting models. RNNs are equipped with an internal memory and can understand sequences more effectively than other types of neural networks. Our work aims to prove the validity of using RNNs with multivariate time series data (related to the Sun’s magnetic fields) to predict solar flares 1 day prior to occurrence. Predicting solar flares by class, 1 day prior to occurrence is also explored. Final analysis indicates successful performance of RNNs when predicting solar flares 1 day prior to occurrence. RNN’s applied to multivariate time series data demonstrate accuracies of 60 percent or more.