Solar Flare Prediction by Real-time Observation Data Analysis of Solar Vector Magnetograms

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K17620
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Astronomy
• Research Institution: National Institute of Information and Communications Technology
• Principal Investigator: Nishizuka Naoto 国立研究開発法人情報通信研究機構, 電磁波研究所宇宙環境研究室, 研究員 (10578933)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 太陽フレア / 予測 / 宇宙天気 / 機械学習 / 衛星画像 / 画像検出 / 深層学習 / ビッグデータ

Outline of Final Research Achievements

Our flare prediction model using machine-learning techniques and big data of solar observation images dramatically improved our skill of the space weather forecast. Whereas the space weather forecasting service has been delivered every day for a long time, we had a difficulty in improving its accuracy. In this research, we applied machine-learning techniques to solar observation data analysis, and we developed a new model to predict solar flares statistically by dealing with the huge amount of data. As a result, we succeeded in improving the prediction accuracy up to 80 %, much better than the human forecasting around 50 %. We also revealed the ranking of features effective for flare prediction. Using our new prediction model, the real-time forecasting operation system of solar flares will be prepared soon.

Free Research Field

太陽物理