Automated prediction of solar flares using neural networks and sunspots associations

Abstract

An automated neural network-based system for predicting solar flares from their associated sunspots and simulated solar cycle is introduced. A sunspot is the cooler region of the Sun's photosphere which, thus, appears dark on the Sun's disc, and a solar flare is sudden, short lived, burst of energy on the Sun's surface, lasting from minutes to hours. The system explores the publicly available solar catalogues from the National Geophysical Data Center to associate sunspots and flares. Size, shape and spot density of relevant sunspots are used as input values, in addition to the values found by the solar activity model introduced by Hathaway. Two outputs are provided: The first is a flare/no flare prediction, while the second is type of the solar flare prediction (X or M type flare). Our system provides 91.7% correct prediction for the possible occurrences and, 88.3% correct prediction for the type of the solar flares. © 2007 Springer-Verlag Berlin Heidelberg.