Synthetic Geomagnetic Field Data Creation for Geomagnetic Disturbance Studies using Time-series Generative Adversarial Networks

Abstract

A key challenge to Geomagnetic Disturbance (GMD) studies is the scarcity of severe geomagnetic field data available to researchers due to its low event occurrence. This study aims to address this challenge by first creating realistic “synthetic” data that represents the geomagnetic field fluctuations caused by recent GMD events. This paper utilizes a machine-learning approach to generate synthetic geomagnetic field data. Specifically, the application and preliminary results of a modified form of the generative adversarial network (GAN) to create time-series synthetic geomagnetic field data of three different severities are described here. The purpose of this paper is to document the first step towards creating severe synthetic geomagnetic field data to advance power system research. Future studies beyond this paper will extend on this work to generate data representing severe GMD storms.