162 Section 3 · Applied use of intelligent computing _________________________________________________________________________________________________________________________________ 3-rd International Conference on Computational Intelligence (ComInt 2015) Taras Shevchenko National University of Kyiv, Cherkasy State Technological University, Cherkasy Institute of Fire Safety named after Heroes of Chernobyl of National University of Civil Protection of Ukraine Kyiv-Cherkasy, Ukraine, May 12-15, 2015 AN OPTIMIZATION APPROACH TO SPACE WEATHER PREDICTION: LYAPUNOV EXPONENTS, PREDICTABILITY, AND REAL-TIME GENETIC ALGORITHMS Vitaliy Yatsenko Space Research Institute of the NASU-SSAU, Kyiv, Ukraine Multi-Input Single-Output (MISO) Nonlinear AutoRegressive Moving Average with eXogenous inputs (NARMAX) models have been derived to forecast the electron fluxes at Geostationary Earth Orbit (GEO). The NARMAX algorithm is able to identify mathematical model for a wide class of nonlinear systems from input-output data. The models employ solar wind parameters as inputs to provide an estimate of the average electron flux for the following day. The identified models are shown to provide a reliable forecast of electron fluxes and are capable of providing real-time warnings of when the electron fluxes will be dangerously high for satellite systems. This report concentrates on applications of global optimization genetic techniques to space weather prediction using satellite experimental data related to solar wind dynamics. New results are: 1) The formulation and optimization method to estimation of Lyapunov exponents from time series of geomagnetic indices; 2) Predictability analysis is conducted for time series using the Kolmogorov-Sinai entropy; 3) An identification algorithm is derived for nonlinear and bilinear dynamical models of geomagnetic indices; 4) A real-time adaptive algorithm for Dst-index prediction is proposed. The above problems have been reduced to nonlinear optimization models with constraints. These models include an estimate of their likely accuracy, a current measure of predictability. Two types of localized Lyapunov exponents based on infinitesimal uncertainty dynamics are investigated to reflect this predictability [1]. New methods developed here are applied to Dst-index prediction, confirming and extending earlier results. The model consists of a bilinear system that has been optimized to be a minimal without degrading the accuracy. Standard methods such as recursive least squares [3], extended least squares, recursive auxiliary variable, and recursive prediction error algorithms have been applied to identifying bilinear systems. Numerical experiments show that our models can be used for operational forecasts of the geomagnetic Dst-index. To solve the problem of geomagnetic indexes prediction in real time it is critical to use high throughput computing to obtain results of prediction just in time. This purpose is achieved with twofold means. The first one suggests sufficiently fast modelling algorithms using a class of discrete dynamic systems which is obtained by perturbation decomposition of correlation function series. The model’s structure and parameter identification is performed by solving mathematical programming problems in rather restricted set of kinds of stable models (may be vast each but efficiently tractable with parallel computing facilities) that gives opportunity to use fast prediction algorithms. Another means is using high performance parallel computing facilities to speed up computation at all stages of analysis-learning-prediction process. It is suggested to use special tools for automated developing efficient parallel programs for mix of standard cluster parallel computer architecture and general purpose graphical processing units. Our tools exploit high-level formal approach to program design and synthesis using algebra-algorithmic specifications and rewriting rules techniques. References 1. Pardalos, P., Yatsenko V. (2006). Optimization approach to the estimation and control of Lyapunov exponents. Journal of Optimization Theory and Applications, 128(1), 29-48. 2. Yatsenko, V., Boyko, Rebennack, N., Pardalos, P.. (2010). Space weather influence on power systems: prediction, risk analysis, and modeling. Energy Syst., (1), 197–207. 3. Yatsenko, V., et. al. (2008). Nonlinear dynamical model for space weather prediction. Ukr. Phys. J., 53(5), 502- 505.