Comparison of Deep Learning Techniques to Model Connections Between Solar Wind and Ground Magnetic Perturbations

Keesee, A.M.; Pinto, V.A.; Coughlan, M.; Lennox, C.; Mahmud, M.S.; Connor, H.K.

Keywords: geomagnetic storms, neural network, space weather, machine learning, LSTM, GIC, ground magnetic field

Abstract

Geomagnetically induced currents (GIC) can drive power outages and damage power grid components while also affecting pipelines and train systems. Developing the ability to predict local GICs is important to protecting infrastructure and limiting the impact of geomagnetic storms on public safety and the economy. While GIC data is not readily available, variations in the magnetic field, dB/dt, measured by ground magnetometers can be used as a proxy for GICs. We are developing a set of neural networks to predict the east and north components of the magnetic field, BE and BN, from which the horizontal component, BH, and its variation in time, dBH/dt, are calculated. We apply two techniques for time series analysis to study the connection of solar wind and interplanetary magnetic field properties obtained from the OMNI dataset to the ground magnetic field perturbations. The analysis techniques include a feed-forward artificial neural network (ANN) and a long-short term memory (LSTM) neural network. Here we present a comparison of both models' performance when predicting the BH component of the Ottawa (OTT) ground magnetometer for the year 2011 and 2015 and then when attempting to reconstruct the time series of BH for two geomagnetic storms that occurred on 5 August 2011 and 17 March 2015.

Más información

Título de la Revista: Frontiers in Astronomy and Space Sciences
Volumen: 7
Número: 550874
Fecha de publicación: 2020
Idioma: English
URL: https://www.frontiersin.org/articles/10.3389/fspas.2020.550874/
DOI:

doi:10.3389/fspas.2020.550874

Notas: Web of Science Core Collection