Predicting Solar Flares Using Machine Learning: Advances and Challenges
Abstract
Robust prediction of solar flares and associated events (coronal mass ejections, solar energetic particles) is of primary importance for the Solar Physics and Space Weather communities. Growing amounts of high-quality observational data together with development and availability of advanced machine learning techniques have initiated machine learning-driven approaches to predict solar flares. I present a brief overview of the current flare prediction methodologies and results, as well as the derived physics knowledge. In particular, the talk covers: 1) statistical approaches as currently implemented at SWPC NOAA and enhanced by the "forecaster-in-the-loop"; 2) feature-based predictions using supervised machine learning techniques and their comparison with SWPC NOAA operational forecasts; and 3) advances in predictions using recently-emerged deep learning approaches. I also discuss common difficulties in cross-comparison of flare prediction attempts such as differences in spacial and temporal scales of the constructed data sets, their class-imbalance ratios and train-test separation. The talk highlights that the machine-learning forecast of solar flares is becoming a reality, but the operational implementation should be properly validated.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMSH34B..05S
- Keywords:
-
- 7536 Solar activity cycle;
- SOLAR PHYSICS;
- ASTROPHYSICS;
- AND ASTRONOMY;
- 7537 Solar and stellar variability;
- SOLAR PHYSICS;
- ASTROPHYSICS;
- AND ASTRONOMY;
- 7544 Stellar interiors and dynamo theory;
- SOLAR PHYSICS;
- ASTROPHYSICS;
- AND ASTRONOMY;
- 7594 Instruments and techniques;
- SOLAR PHYSICS;
- ASTROPHYSICS;
- AND ASTRONOMY