Implementing General Adversarial Networks for Solar Proton Events (SPE) Data Augmentation
Abstract
One of the major space weather challenges is to reliably forecast Solar Proton Events (SPE) to ensure crew and instrument health on long duration space exploration missions. A way to attack this challenge is to implement deep learning techniques to detect SPE events or predict their intensities. In order to implement such techniques, a large amount of data has to be available. However, SPE events are rare, so the related data is rather sparse. In this work, we propose to implement a General Adversarial Network (GAN) that will produce additional synthetic data with characteristics similar to those of the input data. The dataset used is from Richardson et al. (2018) and comprises around a thousand samples of SPE events and non-events at Earth, STEREO-A, and STEREO-B associated with coronal mass ejections. Implementing such a network requires careful analysis and selection of the input parameters and evaluation of the data generated and the network's performance. The synthetic data generated is then combined with real data to train an artificial neural network (ANN) to predict if there will be an SPE event. This study illustrates how space weather phenomena with even modest data available might be candidates for the implementation of machine learning and deep learning techniques.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMSH31E3344C
- 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