NEURAL NETWORKS AND MACHINE LEARNING IN DATA PROCESSING FOR SPACE WEATHER FORECASTING
DOI:
https://doi.org/10.32782/mathematical-modelling/2023-6-2-2Keywords:
neural networks, machine learning, geomagnetic storms, geomagnetic indices, Data ScienceAbstract
To forecast geomagnetic storms, researchers explore not only empirical methods but also analytical approaches, including statistical methods, models based on global-scale physics (magnetohydrodynamic, MHD), methods based on machine learning, or combinations of these. This review is dedicated to developments in artificial intelligence and machine learning aimed at addressing challenges in processing data on geomagnetic activity and predicting space weather. Despite numerous observatories and space missions worldwide collecting and preprocessing data on solar and geomagnetic activity, this process continues to pose challenges to scientists, such as data noise, gaps in time series, and anomalies. All of these are significant hindrances to the development of space weather forecasting, especially in creating real-time predictions, requiring the application of new methods and algorithm development that analyze solar wind speed and coronal mass ejections, ensuring effective forecasting before they reach Earth, which is crucial since solar wind can reach Earth in very short time intervals. Data preprocessing methods include label assigning, handling missing values, and data standardization. It is essential to consider physical phenomena and adapt loss functions for optimal utilization of computer systems in this context. The article mentions two models created for forecasting the Dst index (geomagnetic storms). The first model utilizes a neural network with Long Short-Term Memory (LSTM), trained on data from 2012–2016, achieving an accuracy of 83.47%. The second model, Dst Transformer (DSTT), designed for short-term forecasting, utilizes attention levels and Bayesian inference. DSTT demonstrates high accuracy and addresses two types of uncertainties in the data. Both models are tested and compared with other machine learning methods by the authors. Machine learning enables the identification of complex relationships and forecasting planetary index values in the future, helping to mitigate potential negative impacts of geomagnetic storms on technology and infrastructure. Additionally, it provides people with experience in solving complex scientific problems, which could contribute to new discoveries, inventions, and a better understanding of other physical phenomena in the long run.
References
Wintoft P., Lundstedt H., Eliasson L., Kalla L., Hilgers A. Spacecraft anomaly analysis and prediction system – SAAPS. European Space Agency, (Special Publication) ESA SP. 2001. 476, 169–176.
Smith A. W., Forsyth C., Rae I. J., Garton T. M., Jackman C. M., Bakrania M., Shore R. M., Richardson G. S., Beggan C. D., Heyns M. J., Eastwood J. P., Thomson A. W. P., Johnson, J. M. On the Considerations of Using Near Real Time Data for Space Weather Hazard Forecasting. Space Weather. 2022. 20(7).
Bouriat S., Vandame P., Barthélémy M., Chanussot J. Towards an AI-based understanding of the solar wind: A critical data analysis of ACE data. Frontiers in Astronomy and Space Sciences. 2022. 9(November). 1–18.
Gulati I., Li H., Johnston M., Dlay, S. Classification based Detection of Geomagnetic Storms using LSTM Neural Network. 2022 3rd URSI Atlantic and Asia Pacific Radio Science Meeting AT-AP-RASC 2022. 29 May – 3 June. Gran Canaria, 2022.
Abduallah Y., Wang J. T. L., Bose P., Zhang G., Gerges F., Wang H. Forecasting the Disturbance Storm Time Index with Bayesian Deep Learning. Proceedings of the International Florida Artificial Intelligence Research Society Conference, FLAIRS-35 2022. 15–18 May. Jensen Beach, 2022.
Upendran V., Cheung M. C. M., Hanasoge S., Krishnamurthi G. Solar Wind Prediction Using Deep Learning. Space Weather. 2020. 18(9).
Schmölter E., Berdermann J. Predicting the effects of solar storms on the ionosphere based on a comparison of real-time solar wind data with the best-fitting historical storm event. Atmosphere. 2021. 12(12).
