Deep Learning Prediction of Inter-storm Parameters Using Transformer Convolution Network

Résumé fourni par la source

Abstract The global impact of space weather has created interest in understanding and prediction of geomagnetic storms. Most approaches have considered the prediction of geomagnetic storms based on the time series. In this study, two characteristics of geomagnetic storms will be considered – the time interval and difference in value between two consecutive storms. The Kp index was used as a proxy for geomagnetic activities with values greater than or equal to 4 considered as a storm. The Transformer Convolution Network (TCN) deep learning algorithm was used for the prediction. Optimal model parameters for the two indices were obtained using hyperparameter tuning. Results obtained showed that TCN has good performance in predicting the difference in value but limited performance in time intervals. The limitations of TCN in predicting the inter-storm interval can be attributed to its chaotic nature.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.