Machine Learning Solar Full Disk Flare Operational Forecasting

Solar flare forecasting is an essential component in space environment forecasting. Most of the deep learning flare forecasting models constructed are based on the magnetograms of active regions. Affected by the projection effect, these models can only forecast the active region in the center of the Sun. It is difficult to meet the need of operational flare forecasting of the solar full disk. Based on the traditional solar activity parameters, in this study, the relationships between the magnetic type of the active region, area of the active region, the history of the flare outburst, the 10 cm radio flux and flares from January 1996 to December 2022 were statistically analyzed. By using the fully connected neural network, an operational flare forecasting model for solar full disk active regions was constructed. This model can forecast the eruption of the M-class or above flares of the full solar disk active regions in the next 48 h. The F1 score of the model is 0.4304, the TSS is 0.3689, and the HSS is 0.3906. The model is compared with the deep learning flare forecasting model constructed in the previous work, and the results show that the operational forecasting model constructed in this paper has a better forecasting performance. Meanwhile, in order to explore the influence of the projection effect, the solar full disk active regions flare forecasting model constructed was tested for test data within 30 degrees from the center of the solar disk, within the interval from 30 degrees to 60 degrees, and over 60 degrees, respectively. The results show that the projection effect has little influence on the flare forecast model constructed in this study. The model can be used to forecast flares in the active region of the full solar disk, and provide an effective tool for operational solar flare forecasting.