Date of Award

2024-05-01

Degree Name

Doctor of Philosophy

Department

Computational Science

Advisor(s)

Olac Fuentes

Abstract

Solar active regions are areas on the Sun's surface that have especially strong magnetic fields. Active regions are usually linked to a number of phenomena that can have serious detrimental consequences on technology and, in turn, human life. Examples of these phenomena include solar flares and coronal mass ejections, or CMEs. The precise predictionof solar flares and coronal mass ejections is still an open problem since the fundamental processes underpinning the formation and development of active regions are still not well understood. One key area of research at the intersection of solar physics and artificial intelligence is deriving insights from the available datasets of solar activity that can help us understand solar active regions better. Some machine learning models have been employed to forecast solar flares from a 6-hour to 48-hour time span, thanks to advancements in artificial intelligence. Support Vector Machine (SVM) [5,42], K-Nearest-Neighbor (KNN) [27], Extremely Randomized Trees (ERT) [36], and deep neural network [35] are some of the machine learning models that have been used in forecasting solar flares, but the results are not good. This is due to the models being trained with a specific set of active region parameters and an imbalanced dataset with few positive flare cases. As a result, there is a need to understand space weather and the basis by which these events occur. In this study, we applied a deep learning architecture originally designed for video prediction to predict the changes happening on the Sun in continuous time by using time series Helioseismic and Magnetic Imager data captured by the Solar Dynamics Observatory (SDO) and compared it against a no-change baseline and a regression baseline. In addition, we expanded our study to examine the changes in active regions by incorporating the 3D viewing geometry and the sunâ??s rotation, which helped the models focus on the changes in the active regions. We proposed using log-scale normalization to normalize the data and using the Cascading Convolutional Neural Network to predict the changes in active regions. To improve the performance of the model, we included the gradient information and the Structural Similarity Index in the training of the model by adding them as part of the loss function. In this dissertation, we demonstrated that deep neural networks can be trained to predict changes in active regions. It is our hope that further development of this work will lead to a better understanding of various physical phenomena related to space weather.

Language

en

Provenance

Received from ProQuest

File Size

102 p.

File Format

application/pdf

Rights Holder

Godwill Asare Mensah Mensah

Share

COinS