Solar Flare Prediction with Temporal Deep Learning

Solar Flares are sudden eruptive events of energy from the solar surface that can cause interruptions to various technological systems such as satellite comminications, navigation aids, and onboard electronics. The accurate forecast of solar flares is an important unsolved problem in solar physics. If we can accurately predict flares, we may also be able to completely understand the underlying mechanisms involved in their formation and triggering.

The aim of this project is to utlise a large data set of SHARP parameters, derived from Solar Dynamic Observatory (SDO) measurements to predict the probability of a certain active region flaring within the next 24 hours.

Similar models have been developed using MLP and CNN with limited success. The novel aspect of this work is to utilise temporal convolutional networks (TCNs) to enable the capturing of time-sensitive information of flare dynamics.

If satisfactory performance can be demonstrated this model will be deployed at the Regional Warning Centre of the SANSA in Hermanus.