Solar Flare Prediction with the Hybrid Deep Convolutional Neural Network

Abstract

Abstract We propose a hybrid Convolutional Neural Network (CNN) model (Model 2) and modify a popular CNN model (Model 1) to predict multiclass solar flare occurrence within 24 hr. We collect samples of solar active regions provided by the Space-weather Helioseismic and Magnetic Imager Active Region Patches data from 2010 May to 2018 September. These samples are categorized into four classes (No-flare, C, M, and X), containing 10 separate data sets. Then after training, validating, and testing our models, we compare the results with previous studies in forecast verification metrics with an emphasis on the true skill statistic (TSS). The main results are summarized as follows. (1) This is the first time that the CNN models are used to make multiclass predictions of solar flare. (2) Model 2 has better values of all statistical scores than Model 1 in every class. (3) Model 2 achieves relatively high average scores of TSS = 0.768 for No-flare class, 0.538 for C class, 0.534 for M class, and 0.552 for X class, which are the best results from the existing literatures. (4) Model 2 also can be used to make binary class flare predictions for ≥M-class major flares, and the performance yields a TSS with 0.749 ± 0.079. (5) Model 2 obtains fairly good scores in other metrics for both multiclass flare predictions and ≥M-class major flare predictions. We surmise that some crucial features extracted automatically by our models may have not been excavated before and could provide important clues for studying the mechanism of flare.