Exploring Early Warning of Solar Storms: AI Catches Solar Activity Signals 9.24 Hours Ahead
The research team at NJIT released a machine learning model called EarlyDetect, which can identify precursors to the formation of active regions from solar acoustic activities and magnetic field changes. The best-performing version can capture signals an average of 9.24 hours in advance. This model uses the Transformer architecture to process data from the Helioseismic and Magnetic Imager (HMI) aboard NASA’s Solar Dynamics Observatory. It makes predictions by generating acoustic power maps and combining them with magnetic field measurements. The study shows that when the internal magnetic field of the sun increases, weak traces are left in the sound wave propagation, characterized by a decline in acoustic power followed by continuous changes in intensity and magnetic field. If further verified, satellite communication companies and grid operators may gain more time to adjust their operating strategies and assess risks. However, EarlyDetect has not yet been put into operation.