Modeling of geomagnetic SuperMAG electrojet index by neural networks

The paper solves the problem of SuperMAG electrojet index real-time forecast based on the combination of neural networks and wavelet processing. The scheme of training data generation and architecture of neural networks are illustrated. The quality of modeling of neural networks, trained for different periods, is assessed. Timely forecast of the geomagnetic SuperMAG electrojet index makes it possible to obtain information on expected disturbances in the geomagnetic field. The results of neural network model performance during geomagnetic disturbances are presented. The neural networks show the dependence of forecast quality on the interplanetary magnetic field data.