Method of geomagnetic data analysis based on the combination of wavelet transform with radial basis neural networks

• download

   The present report is devoted to the development of tools intended for the
   analysis of the Earth's magnetic field parameters, the extraction of
   geomagnetic disturbances and estimation of field properties applying
   artificial intelligence and wavelet analysis according to the data
   obtained from ground-based observatories. In this report a method for the
   analysis of the Earth's magnetic field variations (on the example of
   H-vector), based on the combination of wavelet analysis with radial basis
   neural networks, is suggested. This method allows to determine the
   components of field variations, which characterize degree of disturbance
   of field, to study their structure, to classify features of the data and
   to estimate condition of field. On the basis of the data feature
   separation in wavelet space, statistical properties of the process are
   analyzed and informative features are extracted. Extracted features
   determine the radial layer of the network, which specifies belonging of
   the feature to a class.

   The method makes it possible to study the subtle features of geomagnetic
   data structure, and it can be implemented as an automatic tool for rapid
   estimation of the Earth's magnetic field condition. The method has been
   successfully tested on the Earth's magnetic field data obtained from
   Paratunka observatory (Paratunka, Kamchatka region, Far East of Russia).
   The approbation of the method confirmed its effectiveness and allowed to
   extract classification features in the field variations which characterize
   the field disturbance degree.