HR: 0800h
AN: SM41A-1121 [Abstracts]
TI: A genetic programming approach for time-series analysis and prediction in space physics.
AU: * Jorgensen, A M
EM: ajorg@lanl.gov
AF: ISR-4, Los Alamos National Laboratory, MS D448, Los Alamos, NM 87545
United States
AU: Brumby, S P
EM: brumby@lanl.gov
AF: ISR-2, Los Alamos National Laboratory, MS D436, Los Alamos, NM 87545
United States
AU: Henderson, M G
EM: mhenderson@lanl.gov
AF: ISR-1, Los Alamos National Laboratory, MS D466, Los Alamos, NM 87545
United States
AB:
A central theme in space weather prediction is the ability to predict
time-series of relevant quantities, both empirically, and from
physics-based models. Empirical models are often based on educated
guesses, or intuition. The task of finding an empirical relationship
relating quantities can be tedious and time-consuming, especially when
a large number of parameters are involved. Genetic Programming (GP)
provides a method for automating the guesswork, and can in some
instances automatically find functional relationships between data
streams. GP is an evolutionary computation technique which is an
extension of the Genetic Algorithm framework used for function
optimization. In GP an evolutionary algorithm combines elementary
function operators in an attempt to build a function which is able to
reproduce a training example from a set of input data. We will
illustrate how a GP algorithm can be used in space physics by
addressing two relevant topics: The prediction of relativistic
electron fluxes, and prediction of $Dst$.
DE: 2722 Forecasting
DE: 2784 Solar wind/magnetosphere interactions
DE: 2794 Instruments and techniques
SC: SPA-Magnetospheric Physics [SM]
MN: 2004 AGU Fall Meeting