Huazhong Agricultural University (HZAU) has made a breakthrough in space weather forecasting. Li Junpei, a 2021 undergraduate at the College of Informatics, is the first author of a paper published in The Astrophysical Journal Supplement Series, a top astronomy journal. The study, titled A Novel CNN-LSTM Model for Interplanetary Coronal Mass Ejection Detection, introduces an innovative AI-driven method for identifying Interplanetary Coronal Mass Ejections (ICMEs).
ICMEs are among the most violent solar eruptions in the solar system, capable of triggering high-energy particle storms that can disable satellites, paralyze power grids, and endanger astronaut safety. Traditionally, detecting ICMEs relies on manual expert analysis of spacecraft data — a time-consuming and inconsistent process ill-suited for the growing volume of deep-space observations.
To address this, Li's team developed a novel CNN-LSTM hybrid model enhanced with a dynamic loss function. The CNN extracts spatial features, while the LSTM models time-series evolution. The dynamic loss function automatically adjusts weight parameters to counter data imbalance, and the end-to-end training significantly improves computational efficiency.
Using in-situ data from the Wind satellite (1997–2016), the model successfully identified 189 of 230 known ICME events, achieving an F1 score of 81.29 percent — outperforming existing machine learning methods. Notably, detection sensitivity for rare events improved by about 35 percent, offering stronger support for space weather alerts.
The study marks an early step in applying "AI for Science" to space physics.

Characteristic data changes during ICME events. [Photo/news.hzau.edu.cn]