Physics of auroral phenomena : proceedings of the 35th Annual seminar, Apatity, 28 Februaru – 02 March, 2012 / [ed. board: A. G. Yahnin, A. A. Mochalov]. - Апатиты : Издательство Кольского научного центра РАН, 2012. - 187 с. : ил., табл.

between IAI index and indexed parameters introduced additional index ICom. If at this time the value of the indexed parameter or IAI was zero, then the corresponding value ICom assumed to be zero. In assessing the impact of auroral disturbances on the variations in the values IAI the index ICom was calculated as absolute value of quotient CAE/LAI (Fig. 2). In assessing the connection between CMF or CXL parameters with IAI the additional index ICom was taken as 1 in the case of non-zero values of both indices. The displaying o fheliogeophysical disturbance in mid-latitude ionospheric index 3 I *{;.... I I .......i ..| ...t....... .. i In.... 0 100 200 300 Fig. 2. Index IAI values (upper panel) at the station Moscow, index CAE (middle panel), parameter ICom (lower panel) for May 1978. Made calculating of ICom demonstrates reflection of the changes announced heliogeophysical parameters to our proposed index IAI. Overall the total contribution of auroral disturbances, geomagnetic disturbances and disturbances caused by influence of long-wavelength X-rays into midlatitude ionospheric disturbances is about 40%. 3. Reconstructing o f the ionospheric disturbance index values The method of index IAI values reconstructing may be useful to assess the expected level of ionospheric response to the Solar-magnetospheric parameters disturbances. As a tool to reconstructing IAI was used neural networks of the following types: a feed forward network FF (double-layer, 5 neurons in the layer), Elman network ELM (double­ layer, 5 neurons in the layer), neuro network FUZZY (2-5 rules on the input parameter depending on the amount of input data, the form of the rules is "bell"). Neural network experiments were carried out in several stages: 1. Reconstructing the IAI only by history. The correlation coefficient between the real and reconstructed sequences for the feed forward network (FF) was 0.66 for the network Elman - 0.64, and for Fuzzy network - 0.73. Thus the network Fuzzy is the most preferred in this case. 2. Reconstructing the IAI by only one additional parameter. To the input fed parameters one by one XL, MF, AE. Note that in this case the input of the neural network did not serve first derivative of index IAI. Correlation coefficients were low for each of the parameter. 3. Reconstructing the IAI only one additional parameter to include the first derivative of IAI. Compared with the first experiment, the addition of the input parameter XL to feed forward network increases the reconstruction quality at 6%, the parameter MF - 8%, and the parameter AE - 6%. Note that use of additional parameters at feed forward network gives best results whereas in their absence the best result gives network Fuzzy. 4. Reconstructing the IAI for pair combinations of parameters with the inclusion of the IAI first derivative. In this experiment the input submitted by one pair of parameters XL, MF, AE in combination with the IAI first derivative (d (LAI)). In this case the most effective is also feed forward network. Imposition of combinations of two additional parameters increases the average recovery efficiency at 1-2%. Corresponding correlation coefficients between the real and the reconstructed sequence are shown in Table 2. Table 2 Neural network type XL+d(IAI) MF+d(IAI) AE+d(IAI) FF 0.72 0.74 0.72 ELM 0.70 0.68 0.68 FUZZY 0.70 0.69 0.71 Thus by the results of neural network experiments it can be concluded about the possibility of effective recovery which can call short-term forecasting of index IAI values. The preferred type of neural network is a feed forward with the following input parameters: the IAI first derivative and the pair of parameters combination: the geomagnetic field horizontal component plus index AE or geomagnetic field horizontal component plus long 105 1978 ______i__________» » ,„ SCO 600 700 800 >.hours