Prediction of magnetospheric parameters using artificial neural networks
Doctor of Philosophy thesis
Artificial neural network models have been developed that provide the magnetospheric parameters Dst, polar cap potential and the midnight equatorward boundary of diffuse aurora. Layered feedforward neural networks have successfully learned the relationship between the solar wind and the magnetospheric parameters using supervised back-propagation training. All models have achieved a higher prediction accuracy than the existing empirical or statistical models. These models are applied to the prediction of the parameters, which will then be used by the Rice Magnetospheric Specification and Forecast Model (MSFM). The neural network models are able to forecast the magnetospheric parameters 30 to 60 minutes ahead using the information from a solar wind monitor spacecraft. With the forecast values, the MSFM will be able to forecast particle fluxes in the inner magnetosphere. The MSFM is applied to the April 1988 magnetic storm for the forecast capability test. The neural network modeling, the comparison of the prediction accuracy with other methods and the result of the MSFM forecast capability test are presented.
Statistics; Physics; Artificial intelligence; Computer science