Deep learning teachology for the prediction of solar flares from GOES data

Predicting solar storms from real-time satellites data is extremely important for the protection of variousaviation, power and communication infrastructures. In this paper deep learning technologyis applied, for the first time, tothe real-time prediction of solar flares by analyzing the x-ray flux (1-minute cadence) time-series data from the satellite GOES (Geostationary Operational Environmental Satellite). The prediction system introduced here consists of 2units. The first converts GOES data to Markov Transition Field (MTF) images. An unsupervised feature learning algorithm and the prediction, flare or no-flare, are implemented by a Convolutional Neural Network in the second unit. Severalevaluation metrics, as required by space weather specialists, are applied to evaluate the performance of the system.