CNN-Based Deep Learning in Solar Wind Forecasting

Abstract

This article implements a Convolutional Neural Network (CNN)-based deep learning model for solar-wind prediction. Images from the Atmospheric Imaging Assembly (AIA) at 193Ȧ wavelength are used for training. Solar-wind speed is taken from the Advanced Composition Explorer (ACE) located at the Lagrangian L1 point. The proposed CNN architecture is designed from scratch for training with four years' data. The solar-wind has been ballistically traced back to the Sun assuming a constant speed during propagation, to obtain the corresponding coronal intensity data from AIA images. This forecasting scheme can predict the solar-wind speed well with a RMSE of 76.3 km\s and an overall correlation coefficient of 0.57 for the year 2018, while significantly outperforming benchmark models. The threat score for the model is around 0.46 in identifying the HSEs with zero false alarms.

arXiv comment21 pages,13 figures. Accepted for publication in Solar Physics. After published, it will be available at https://www.springer.com/journal/11207

References in corpus (11)

• Adam: A Method for Stochastic Optimization
• Scikit-learn: Machine Learning in Python
• Array Programming with NumPy
• Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning
• Coronal Holes
• Application of Deep Convolutional Neural Networks for Detecting Extreme Weather in Climate Datasets
• A Machine Learning Dataset Prepared From the NASA Solar Dynamics Observatory Mission
• Solar wind prediction using deep learning
• Real-time solar wind prediction based on SDO/AIA coronal hole data
• SDO Observations of Magnetic Reconnection at Coronal Hole Boundaries
• Using gradient boosting regression to improve ambient solar wind model predictions

Cited by in corpus (3)

• Outlier-based Autism Detection using Longitudinal Structural MRI
• Forecasting High-Speed Solar Wind Streams from Solar Images
• Scope and limitations of ad hoc neural network reconstructions of solar wind parameters