HR: 0830h
AN: SM51A-09    [Abstracts]
TI: Geomagnetic Activity Forecasting Using Self-Learning Algorithms: Application in Space Weather Studies
AU: * Khalil, A F
EM: akhalil@cc.usu.edu
AF: Utah State University, Department of Civil and Environmental Engineering 4110 Old Main Hill, Logan, UT 84322-4110 United States
AU: Barakat, A R
EM: arb@cc.usu.edu
AF: Utah State University, Center for Atmospheric and Space Sciences 4405 Old Main Hill, Logan, UT 84322-4405 United States
AU: McKee, M
EM: MacMcKee@usu.edu
AF: Utah State University, Department of Civil and Environmental Engineering 4110 Old Main Hill, Logan, UT 84322-4110 United States
AB: The ability to forecast the geomagnetic activities is becoming more important as human activity in space becomes more prevalent. For example, early warning of geomagnetic storms could help mitigate their harmful effects on space electronics and on electrical power lines. Moreover, recently developed space weather algorithms that utilize physics-based models require future values of Kp as an input in order to forecast the ionospheric behavior. Computational learning theory and data-driven modeling techniques are new and rapidly expanding areas of research that aim at developing efficient learning algorithms. Here we compare self-learning algorithms regarding their abilities to forecast the level of geomagnetic activities, as represented by Kp. In particular, we consider the following algorithms: artificial neural networks, locally weighted projection regression, support vector machines, and relevance vector machines. Different parameters are considered such as: (1) length of forecasting time, (2) type and size of input data, and (3) training set size. These learning machines are compared regarding their generalization capabilities and structure reliabilities. The relative strengths and limitations of these algorithms will be presented.
DE: 2447 Modeling and forecasting
DE: 2722 Forecasting
DE: 2788 Storms and substorms
DE: 2794 Instruments and techniques
SC: SPA-Magnetospheric Physics [SM]
MN: 2005 Joint Assembly