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.

О.M. Barkhatova, S.E. Revunov The main results o f linear correlations are: 1. The Solar wind velocity and its density have greater influence on a ring current (SYM, ASY) in comparison with auroral electrojets. It is expressed in sharp increase o f correlation at the fixed delay. The correlation analysis allows to establish the typical advancing times of PSW and IMF By component displaying in symmetric and asymmetric currents. They are about 70-80 minutes. The consideration of received correlation connections between SYM and ASY indices and PSW and IMF parameters gives that these parameters become apparent in ring current symmetric part more often in comparison with its asymmetric part. 2. IMF components By and Bz have the greatest influence on auroral electrojets. It is marked for geomagnetic storms o f any intensity. Further on importance follow the Solar wind density and then the Solar wind velocity. For storms of strong intensity the Solar wind velocity influence is greater than for storms of average intensity and weak storms. Typical advancing times of PSW and IMF displaying in electrojets are 70-80 minutes. 4. The search of nonlinear connections and neural network reconstruction In all series of completed neural network experiments the PSW and IMF values have been used as training sequence, and as test sequence - values of magnetospheric and ionospheric currents indices. For this research the Elman neural network has been chosen which contained 4 hidden layers with 6 neurons in everyone. Experience has shown that such network configuration is optimum because the increase of layers quantity or neurons in layer frequently leads to its overload. The performance o f neural network experiments for each pair of indices was carried out in such sequence: 1. For chosen training parameter (PSW or IMF) typical advancing times of the reconstructed parameter (AU, AL, SYM, ASY) were proposed on the basis o f linear correlations. This time was equaled to 80 minutes and takes into account time of PSW and IMF carrying to magnetosphere boundary. 2. Neural network has been trained on one o f thirty events. On network input feed values of the training parameter and its first derivative taking into account established advancing time. 3. Trained neural network used for the reconstruction of others 29 events. The correlation coefficient between real and reconstructed sequences was calculated for an estimation of reconstruction efficiency. Completed neural network experiments show that successful reconstruction of ring current and auroral electrojets intensity indices on the PSW and IMF parameters data is possible. Neural network reconstruction was considered successful if correlation between real and reconstructed values were in an interval from 0.4 up to 0.9. Fig. 1 shows examples of ring current and auroral electrojets intensity indices reconstruction on the PSW and IMF data. V-AU V-AL By-AU By -AL Fig. 1. Examples of the neural network reconstruction of auroral electrojets and ring current intensity indices on the PSW and IMF data. The solid line shows real values o f indices and dashed line - the values of indices reconstructed by neural network. 2 2