Physics of auroral phenomena : proceedings of the 40th annual seminar, Apatity, 13-17 March, 2017 / [ed. board: N. V. Semenova, A. G. Yahnin]. - Апатиты : Издательство Кольского научного центра РАН, 2017. - 143 с. : ил., табл.

N.A. Barkhatov et al. Input layer Hidden layer Output layer A_____ A >’*(/) xa> “O c: 1 . 2 . I 0.4 03 о сi 0.8'- \ p y , 0.0 R=0.83 PE=79% Figure 1 500 1500 2500 3500 time, minutes Figure 2 Thus, it was found that the work of neural networks demonstrates the effect o f controlling high-latitude geomagnetic activity by the parameters of the magnetic field of the cloud. In experiments to effectively restore the AL index, the need for 90 minutes of the prehistory of the combination of MMP components was shown. It indirectly indicates the dependence of the dynamics of the substorm activity on the structure of the large-scale configuration of the magnetic field of the cloud in the Earth's orbit. The created redundancy o f the input array stabilizes the ANN, which by the high quality of synthesis of the amplitude values at the output is demonstrated. In Fig. 2 shows an example of comparing real values of AL-index (gray curve) and simulated ANN (black curve) on June 15, 2005. The abscissa shows the time in minutes, and the ordinate shows the normalized values of the AL index. The conclusions obtained in the formulation of numerical ANN experiments on the recovery of the AL index indicate the possibility of using ANN as a magnetic cloud detector. The created neural network, using IMF parameters for intervals corresponding to magnetic clouds, is capable of successfully generating an AL index dynamics comparable to the actual situation. Further, we check the capabilities of the ANN on the data intervals of group 3 (70 cases) corresponding to isolated magnetospheric substorms, which are certainly not associated by magnetic clouds. As the results show, if data intervals not corresponding to magnetic clouds (group 3) are fed to ANN inputs, then the quality of recovery of dynamics for AL index drops sharply, even if the data o f the same group (R ~ 0.3%, PE ~ 5%). This allows us to state that the network architecture found to the problems of identifying solar plasma streams with magnetic clouds can be applied. The main conclusions of the study can be formulated as follows: 1. Using By and Bz IMF components that correspond to a magnetic cloud as input parameters of the neural network model taking into account 90 minutes of prehistory is enough to restore the sequence of the AL index. 2. The Elman ANN architecture with an external feedback loop demonstrates a satisfactory recovery of the AL index. 3. The developed model of recovery of the AL index in problems of identifying solar plasma streams with magnetic clouds. Only data intervals corresponding to magnetic clouds can successfully generate an AL index comparable to the actual situation at the output o f the neural network model. This is verified on the data intervals of group 3, which, according to the indications of the AL index, correspond to isolated magnetospheric substorms, which are certainly not associated with magnetic clouds. The completed research researches showed that for the recovery of the AL index sequence with efficiency up to 80% it is sufficient to use the By and Bz IMF components taking into account their 90 minutes of prehistory as input parameters of the neural network model. This means that, during periods of interaction of the Earth's magnetosphere with magnetic clouds, there is a close nonlinear relationship between the level of magnetic activity in high latitudes and dynamics of By and Bz IMF components. The created neural network model with high efficiency to restore both separate substorms and substorms caused by slow magnetic clouds [4] can be used. Acknow ledgments. This work was supported by grant RFBR №16-05-00608, №16-35-00084 and State Task of Minobmauki RF № 5.5898.2017/8.9. References [1] Henderson M.G., Reeves G.D., Belian R.D., Murphree J.SD. Observations of magnetosphericsubstorms occurring with no apparent solar wind/IMF trigger // J. Geophys. Res. V. 101. No. A5. P. 10773-10792. doi 10.1029/95JA00186.1996. [2] Barkhatov N.A., Vorob’evb V.G., Revunov S.E., Yagodkina O.I., Vinogradov A.B. Demonstration of reflection of solar wind dynamics parameters during formation substoimactivity using a predictive tool // Proc. of 39 Annual Seminar «Physics ofAuroral Phenomena». PGI. Apatity. P. 27-30. 2016 (in Russian) [3] Barkhatov N.A., Revunov S.E., Uryadov V.P. Forecasting of the critical frequency of the ionosphere F2 layer by the method of artificial neural networks // Int. J. Geomagn. Aeron. GI2010, doi 10.1029/2004GI000065. 2004. [4] N.A. Barkhatov, V.G. Vorob’evb, S.E. Revunov, and O.I. Yagodkina Effect of Solar Dynamics Parameters on the Formation of SubstormActivity // Geomagnetism and Aeronomy, 2017, Vol. 57, No. 3, pp. 251-256. © Pleiades Publishing, Ltd., 2017 75