Physics of auroral phenomena : proceedings of the 33rd Annual seminar, Apatity, 02 - 05 March, 2010 / [ed.: A.G. Yahnin, A. A. Mochalov]. - Апатиты : Издательство Кольского научного центра РАН, 2011. - 206 с. : ил.

N.A. Barkhatov, S.E. Revunov 4) Forecasting Dst index for real-time data from the ACE spacecraft for the next hour using Elman ANN [http://rwc.lund.irf.se/rwc/dst/models.php] . Advantages: free provision o f models for calculating Dst in two formats - in the format o f MATLAB and the format of the Java application. Providing detailed description of the model. Providing history for prediction within 1 week. Providing data in digital form. 5) Mapping the distribution of energy for injected particles in the polar regions using Dst index value [http://climatology.space.swri.edu/tedimages.html] . Advantages: choice for mapping the range of particle energies, the ability to select source that provides data on Dst index, ability to manually input the values of Dst index. Provide descriptions of the model used. Providing data in digital form. The main purpose of creating our software and computing complex neural networks series of numerical experiments to find a nonlinear relation between geomagnetic indices among themselves and with the near Earth space parameters and for the complex classification o f space weather events are provided. Particular attention to the development of application architecture was paid. The proposed open modular architecture allows you to easily modify it to solve different problems. The core complex a set of neural network algorithms includes: artificial neural network algorithm with Elman back propagation network, feed forward network algorithm, fuzzy logic network, Kohonen layer classification network. Selected algorithms o f artificial neural networks in a wide range o f tasks such as forecasting and recovering of numerical series with history of process and associated parameters, clustering and division data on feature set haye proven themselves. 2. Application interface Software and computer system for predicting space weather phenomena for work through the Internet is currently available. Service for our complex at Research Laboratory of Physics of solar-terrestrial relationships NNSPU is posted. Its web-interface available on address http://spacelabnnov.110mb.com. For free access to the system select link “Forecast parameters of solar-terrestrial relationships”, and then - “Neural network modeling”. The application interface to fme-tuning the parameters o f the experiment allows. The user can select a template architecture and learning algorithm o f neural networks, input parameters for training and testing, spline processing parameters for input data, data formats, number o f neurons in the layers of the network, the number of hidden layers, data packets for training and testing, modes of processing and presentation of results. The.beginning of the training process with selection of ANN input and target data in main window of the program is consists. The result of network work is the plot with results o f testing. For simplification the setting of repeating experiments the application allows last trained neural network to use. In this case training is not done: from a previously saved file parameters of experiment are loaded. Previously trained network as a filter for input parameters get through and output parameters with the test event are compared. 3. Used data For sure o f the greater flexibility of the program work with a database in the form of files is stipulated. The created complex for 30 simultaneously processing streams of events in the input data is designed. Database minute numeric data for 30 geomagnetic storms in the period from 2000 to 2003 is contains. In this set of parameters of the Solar wind, Interplanetary Magnetic Field (from spacecraft “ACE”) and geomagnetic indices (Dst, SYM, ASY, AU, AL) are included. These data from the site at http://cdaweb.gsfc.nasa.gov are received. Is it necessary discrete data can be modified using the spline. 4. Statement o f experiments through the web- interface With access to the system the user enters a query to the main page o f calculation. This request contains sections. Their consequent fill will perform neuromodelling between the given parameters and get the final product - artificial neural network answer. Appearance o f the main page of request of calculation in browser Internet Explorer window is presented in a Fig. 1.