We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
The Short‐Time Prediction of the Energetic Electron Flux in the Planetary Radiation Belt Based on Stacking Ensemble‐Learning Algorithm.
- Authors
Tang, Rongxin; Tao, Yuhao; Li, Jiahao; Chen, Zhou; Deng, Xiaohua; Li, Haimeng
- Abstract
High energy electrons in planetary radiation belts are a major threat to satellites and communications in deep space applications. In order to predict the variations of energetic electron fluxes for different energy channels, we proposed a new ensemble machine leaning model for differential electron flux from 30 keV to 4 MeV in the Earth's radiation belts based on the RBSP‐A observation data from March 2013 to December 2017. The deep neural network (DNN), the convolutional neural network (CNN), the combination of CNN and DNN (CNN&DNN), the linear regression (LR), and the light gradient boosting machine (LightGBM) are among the machine learning models chosen. We carefully compared the electron flux predictions for 20 energy levels and all five models can present valid short‐time flux forecasts. The DNN model has the poorest result. The LR model is good for short‐term forecasting but not so good for long‐term forecasting. The LightGBM ensemble model is highly stable, and it has always outperformed other independent models in terms of forecast accuracy. Then the comparison by adding AE and SYM‐H indexes is given and the influence of geomagnetic activity conditions can be negligible for this short‐time prediction. Furthermore, we applied these five models on Saturn and finally got very similar prediction results. Our results will be significantly useful in instrument designs and system control of future scientific satellites in deep space explorations. Plain Language Summary: With the development of AI hardware technology, it can be predicted that those smart chips will soon be applied to various observation instruments carried by space exploration satellites. However, based on the requirements of energy limits and physical dimensions in satellites, the smart chips used in the instruments can not have very powerful computing power, which may not be effective for current long‐time space weather predictions with many large‐scale parameters. In the present study, we present a short‐time electron flux forecast in planetary radiation belts whose experiments only consider the historical flux data and local parameters. We used five different machine learning models to give the short‐time prediction of the electron differential fluxes for all 20 energy levels from tens of keV to several MeV based on RBSP‐A observations. We also verified this ensemble model on the electron flux forecast near Saturn's equatorial plane by using CASSINI/MIMI data. Our work give a new insight in the space weather applications based on machine learning, which will be critical for practical applications to the instrument designs and system operations of new scientific satellites in the space radiation environment. Key Points: We present a short‐time prediction of the electron flux in the planetary radiation belt based on five different machine learning methodsWe compared the electron flux predictions from 30 keV to 4 MeV and all five models can present valid short‐time flux forecastsFurther experiments on Saturn also show the validation of the ensemble model
- Subjects
PLANETARY atmospheres; CONVOLUTIONAL neural networks; MACHINE learning; REGRESSION analysis; SPACE exploration
- Publication
Space Weather: The International Journal of Research & Applications, 2022, Vol 20, Issue 2, p1
- ISSN
1539-4956
- Publication type
Article
- DOI
10.1029/2021SW002969