Solar Flare Forecasting Based on Fusion Model

Solar flare prediction is an important subject in the field of space weather, and deep learning technology has greatly promoted the development of this subject. In this study, we propose a novel solar flare forecasting model integrating Deep Residual Network (ResNet) and Support Vector Machine (SVM) for both ≥ C-class (C, M, and X classes) and ≥ M-class (M and X classes) flares. Samples of magnetogram from May 1, 2010 to September 13, 2018 are collected from Space-weather Helioseismic and Magnetic Imager (HMI) Active Region Patches. We then obtain 7 independent data set based on cross-validation method, and utilize 5 metrics to evaluate our fusion model based on intermediate-output extracted by ResNet and SVM with the Gaussian kernel function. The results show that the primary metric true skill statistics (TSS) achieves a value of 0.708 ± 0.027 for ≥ C-class prediction, and of 0.758 ± 0.042 for ≥ M-class prediction, which have significant advantage over previous studies. Meanwhile, the metrics of our fusion model on 7 datasets show strong stability and robustness. This suggests that a fusion model integrating an excellent baseline network and SVM can achieve better performance in solar flare prediction. In addition, the performance impact of architectural innovation in our fusion model is also discussed.