Real-Time Machine Learning Enabled Low-Cost Magnetometer System

Geomagnetically Induced Currents (GICs) are one of the most hazardous effects of space weather. The rate of change in ground horizontal magnetic component $dB_{H}/dt$ is used as a proxy measure for GIC. In order to monitor and predict $dB_{H}/dt$, ground-based fluxgate magnetometers are used. However, base-line correction is crucial before such magnetometer data can be utilized. In this paper, a low-cost Machine Learning (ML) enabled magnetometer system has been implemented to perform real-time baseline correction of magnetometer data. The predicted geomagnetic components are then used to derive a forecast for $dB_{H}/dt$. Two different ML models were deployed, and their real-time and offline prediction accuracy were examined. The localized peaks of the predicted $dB_{H}/dt$ are further validated using binary event analysis.