Predicting Solar Flares Using a Convolutional Neural Network with Extreme-ultraviolet Images

Solar flares, as one of the most intense solar eruptive activities, can bring a destructive impact on the near-Earth space environment and technological infrastructure. Therefore, accurate and real-time forecast of the occurrence of flares is essential to reduce the potential damage to space facilit...

Full description

Saved in:
Bibliographic Details
Main Authors: Yun Yang, Yi Wei Ni, P. F. Chen, Xue Shang Feng
Format: Article
Language:English
Published: IOP Publishing 2025-01-01
Series:The Astrophysical Journal
Subjects:
Online Access:https://doi.org/10.3847/1538-4357/adc88e
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Solar flares, as one of the most intense solar eruptive activities, can bring a destructive impact on the near-Earth space environment and technological infrastructure. Therefore, accurate and real-time forecast of the occurrence of flares is essential to reduce the potential damage to space facilities and navigation systems. Several types of flare forecasting methods have been proposed, e.g., the statistical methods, numerical methods, traditional machine learning, and the recently developed deep leaning. In this paper, we apply one of the deep learning methods, the convolutional neural network ResNet-18, to the tricolor (94, 193, and 304 Å) full-disk extreme-ultraviolet images observed by the Solar Dynamics Observatory Atmospheric Imaging Assembly from 2012 to 2022 with 5 × 24 × 365 × 11 images and 10026 flares ≥ C-class in total to build a solar flare forecasting model to predict whether a flare ≥ C-class would occur within the next m hr ( m = 6, 12, 24, 48, and 72). The values of most performance evaluation metrics scores are over 0.8, with some even exceeding 0.9. The main advantage of this model lies in that it can effectively leverage the historical and spatial information of the source active regions as well as the interaction information from the surrounding active regions or other structures. This information, which cannot be provided by the data from a single active region alone, is demonstrated to be crucial in predicting solar flares. The results show that our prediction model performs well in both short-term and long-term forecasting.
ISSN:1538-4357