ИСТИНА |
Войти в систему Регистрация |
|
ИПМех РАН |
||
The Earth’s magnetosphere is a complex dynamic system whose state is hard to predict due to its dependence on solar wind, interplanetary magnetic field, and also on its own recent history. Artificial neural networks are able to take into account all these factors and to construct a non-linear function mapping them to future values of geomagnetic Dst index, which is usually used to characterize the degree of disturbance of the Earth’s magnetosphere. This study report use of ensemble approach and stacked generalization to improve the quality of neural network prediction of Dst index with prediction horizon of up to 12 hours.