A solar wind-based model of geomagnetic field fluctuations at a mid-latitude station

Anomalous quasi-DC currents known as geomagnetically induced currents (GIC), produced in electric power network infrastructure during geomagnetic storms, pose a risk to reliable power transmission and network integrity. The prediction of a geomagnetic field-derived proxy to GIC provides an attractive mitigation technique that does not require changes to network hardware. In this paper we present the development of two artificial neural network based models tasked with predicting variations in the X (northward) and Y (eastward) components of the geomagnetic field at Hermanus, South Africa, with only solar wind plasma and interplanetary magnetic field (IMF) parameters as input. The models are developed by iteratively selecting the best set of solar wind parameters to predict the fluctuations in X and Y. To predict the variation in X  , IMF magnitude, solar wind speed, fluctuation in solar wind proton density and a IMF-BZBZ derived parameter are selected. To predict the variation in Y  , IMF-BZBZ, solar wind speed, and fluctuation in IMF magnitude are selected. The difference between the sets of selected input parameters are explained by the dependence of eastward perturbations in geomagnetic field at middle latitudes on field aligned currents. Model performance is evaluated during three storms in 2012. The onset and main phases of storms are fairly accurately predicted, but in cases where prolonged southward IMF coincides with solar wind parameters that are slowly varying the model fails to predict the observed fluctuations.