CME Arrival Time Prediction Using Convolutional Neural Network

• Published in: The Astrophysical journal Vol. 881; no. 1; pp. 15 - 24
• Main Authors: Wang, Yimin, Liu, Jiajia, Jiang, Ye, Erdélyi, Robert
• Format: Journal Article
• Language: English
• Published: Philadelphia The American Astronomical Society 10.08.2019; IOP Publishing
• Subjects: Artificial neural networks; Astrophysics; Coronal mass ejection; Neural networks; Prediction models; Regression analysis; Regression models; solar-terrestrial relations; Sun: coronal mass ejections (CMEs); techniques: image processing; Transit time; White light
• DOI: 10.3847/1538-4357/ab2b3e
• ISSN: 0004-637X

Fast and accurate prediction of the arrival time of coronal mass ejections (CMEs) at Earth is vital to minimize hazards caused by CMEs. In this paper, we use a deep-learning framework, i.e., a convolutional neural network (CNN) regression model, to analyze transit times from the Sun to Earth of 223 geoeffective CME events observed in the past 30 yr. 90% of them were used to build the prediction model, and the rest 10% have been used for test purpose. Unlike previous studies on this topic, our proposed CNN regression model does not require manually selected features for model training, it does not need time spent on feature collection, and it can deliver predictions without deeper expert knowledge. The only input to our CNN regression model is the instances of the white-light observations of CMEs. The mean absolute error of the constructed CNN regression model is about 12.4 hr, which is comparable to the average performance of the previous studies on this subject. As more CME data become available, we expect the CNN regression model will reveal better results.