Solar flares are immense energy explosions in the solar atmosphere and severely influence space weather. So, forecasting solar flare eruptions is extremely important. Spatial distribution and evolution of active region (AR) magnetic fields are closely related to flare eruptions. In this paper, we simultaneously utilized the two characteristics to build two flare-forecast models using three-dimensional convolutional neural networks (3D CNNs). The two models forecast whether an AR would erupt a ≥C- or ≥M-class flare within the next 24 hr, respectively. The magnetogram sequences provided by the Space-weather Helioseismic and Magnetic Imager Active Region Patches are selected to train our models. We used several performance metrics, such as true skill statistics (TSS), to evaluate our models. The TSS scores of the ≥C and ≥M models reach 0.756 and 0.826, respectively, indicating that our models have superior forecast performance. We used the the gradient-weighted class activation mapping (Grad-CAM) method to visually explain our flare-forecast models. The Grad-CAM illustrates that the 3D CNNs may extract the spatial distribution and evolution of AR magnetic fields simultaneously for flare forecasts. So, the 3D CNN method is valid for flare forecasts, and it utilizes the characteristics related to flare eruptions.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.