Prof. DENG Linhua from Yunnan Observatories of the Chinese Academy of Sciences with collaborators (Prof. WANG Feng from Guangzhou University and Prof. FENG Song from Kunming University of Science and Technology), proposed a fine-grained model for solar flare forecasting based on the Hybrid Convolutional Neural Networks (CNN), they found that it is valuable to separately model the flare forecasts in the rising and declining phases of a solar cycle.

This research, accepted by The Astrophysical Journal, found that the flare forecasting models developed have excellent accuracy and stability. Significantly, the accuracy of the two models for the rising and declining phases is more advantageous, which provides a more reliable means to carry out space weather forecasting in the later stage.

Solar flares are one of the most intense solar activities, manifested mainly in the sudden enhancement of the radiation flux from the radio band up to the X-rays. Previous studies showed that solar flares have a significant impact on Earth’s environment and human activities. Therefore, the study of solar flares, especially forecasting flare eruptions, has become a research hotspot and an essential element of space weather forecasting.

With the rapid development of computer technology, machine learning technology was applied to the forecasting of solar flares and played an important role. Previous studies for forecasting solar eruptions have yielded valuable results. However, there are still some shortcomings in the current research. Firstly, the number of samples is insufficient, especially the X-class flare samples. Secondly, all the current research work has been done for the whole solar cycle, and there is a lack of finer-grained research, such as the influence of rising and falling phases on flare forecast.

In this study, they propose and validate a more stable hybrid CNN model for forecasting flares within 24 hours in a full solar cycle. Based on this model, they further investigate the performance of fine-grained models on rising phase (Mrp) and declining phases (Mdp) of solar cycle 24, respectively.

They found that the true skill statistics (TSS) scores of both Mrp and Mdp for No-flare, C-class, M-class, and X-class were greatly improved than the previous studies. Combined with the forecast result of Mrp, it proved that investigating the modeling for the rising phase and the declining phase is an effective method to improve the accuracy of forecasting solar flare.

Contact:

Linhua Deng, Yunnan Observatories, Chinese Academy of Sciences

lhdeng@ynao.ac.cn