S1-P11


NEURAL NETWORK PREDICTION OF GEOSYNCHRONOUS RELATIVISTIC
ELECTRON FLUX FROM SOLAR WIND DATA


P. WINTOFT and H. Lundstedt


Swedish Institute of Space Physics, Solar-Terrestrial
Physics Division, Lund



E-mail: peter@irfl.lu.se



Electrons with MeV energies are believed to cause deep
dielectric charging of spacecraft at geosynchronous orbit. The charge
build-up and the subsequent discharge may cause damage to the
spacecraft. In this work we show how hourly averages of the > 0.6 MeV
and > 2 MeV electron flux can be predicted from solar wind data several
hours in advance. Specifically we use the OMNI and ACE solar wind data
sets, and the GOES-08 and -10 electron data. As the variance of the
electron flux is very different for different local time sectors the
model must take this into account. Thus, the model consists of expert
neural networks each predicting the electron flux for a certain local
time. Even though the GOES satellite only samples the electron flux at
one specific local time sector at any given time, the model can
generalize so that predictions can be made for any local time sector.
The predictions are compared with measured data for a few test periods.