The Rule of Artificial Neural Network Algorithm in Geomagnetic Storms Prediction

Abstract

While relativistic electrons can completely destroy a spacecraft when the solar wind-magnetospheric interactions are enhanced, the Dst index is considered to be an indicator of any geomagnetic storm. The more negative the Dst index values, the stronger the magnetic storm.   Every relativistic electron event was associated with a magnetic storm, but, magnetic storms could occur without appreciable enhancement of the relativistic electron fluxes. The problem thus arises, which one should be predicted:  the Dst index or relativistic electron enhancements (REE), in order to be more logic? and which is more effective for prediction: the use of statistical relationships or Artificial Neural Networks? Reproduction (or simulation) of the Dst index using a neural network algorithm would solve the problem.

An Artificial Neural Network Algorithm was adopted in the present study for the reproduction of the Dst index of geomagnetic storms having the training concept “Train to Gain” in mind.  The ANN was well trained using a data set of 37 storms of different intensities as input to the network. A well trained ANN would yield an extremely good correlation between the measured Dst and the predicted Dst.

The applied ANN algorithm in the present study shows an excellent performance. About 97% of the Dst have been reproduced, at least, for both the main and recovery phases. Efficient forecast of the oncoming relativistic electron flux enhancements (REE) can thus - under certain conditions - be issued.

Keywords: Geomagnetic storms, Geosynchronous orbit, Solar cycle-23, Dst index, Relativistic Electron Enhancement, Artificial Neural Network.