An Application of the Deep Belief Networks to the prediction of massive Solar Flare Occurrence Using SDO/HMI Data

A solar flare is an eruption on the sun that happens when energy stored in the twisted magnetic field of solar active regions is suddenly released. The physical mechanisms for these flares are still not fully understood. This paper presents a novel method using the Deep Belief Networks (DBNs) for the prediction of the solar flares occurrence. The input data for the verification of this system are SDO/HMI Intensitygram and Magnetogram images (from 2012 Jan to 2013 December with 1 h cadence). The model outputs are "Flare or No-Flare" of major flare occurrence (M, and X-class flares). To obtain a large amount of observational data to train and test the performance of the DBN model used in this system, a dataset is created from the sunspots groups extracted from SDO HMI Intensitygram images from November 1, 2010, to December 31, 2011, and January 1, 2014, to June 1, 2017. Furthermore, we compared the results obtained from the complete suggested system with those of three previous flare forecast models using several statistical metrics. The major findings of this study are that the results of the proposed model are slightly higher in Heidke skill score and true skill statistics, and better in false alarm ratio (FAR) than the previous solar flare prediction systems.