CAS Institute of Automation and National Astronomical Observatories Collaborate on Stellar Flare Prediction Model

The Chinese Academy of Sciences' Institute of Automation has collaborated with the National Astronomical Observatories to develop FLARE (Forecasting Light-curve-based Astronomical Records via Features Ensemble), a comprehensive model designed to predict stellar flares. Stellar flares, which involve the rapid release of magnetic energy from a star’s atmosphere, are crucial for understanding stellar structure, evolution, magnetic activity, and even the search for habitable exoplanets and extraterrestrial life. However, the limited number of observed flare samples has hindered deeper research. Accurate prediction of stellar flares is thus a significant challenge in astronomy. Unlike solar flares, which can be relatively easily predicted due to the wealth of data available, stellar flares rely primarily on light curve analysis. Light curves often suffer from data gaps and exhibit significant variations among different stars and even for the same star over different periods, making flare prediction particularly complex. FLARE addresses these challenges by integrating stellar physical properties, such as age, rotation speed, and mass, with historical flare records. The model employs a unique soft prompt module (SPM) and a residual record fusion module (RRFM) to effectively combine this data, enhancing its ability to extract meaningful features from light curves. This integration significantly improves the accuracy of flare predictions. In their experiments, the research team used high-precision light curve data from 7160 stars. FLARE outperformed several other models, including multilayer perceptrons (MLPs), recurrent neural networks (RNNs), convolutional neural networks (CNNs), graph neural networks (GNNs), transformers, and pre-trained language models adapted for time series analysis. FLARE achieved an accuracy rate of over 70%, excelling in multiple evaluation metrics such as F1 score, recall, and precision. Moreover, FLARE demonstrates strong adaptability, capable of accurately predicting flares based on the diverse patterns of light curves from different stars. Even for the same star, where light curve patterns vary over time, the model can still make precise predictions. This robust performance makes FLARE a valuable tool for astronomers, potentially aiding in the discovery of more cosmic mysteries. The development and success of FLARE highlight the enormous potential of combining artificial intelligence with scientific research, particularly in the field of astronomy. By leveraging AI, researchers can process and analyze vast amounts of astronomical data more efficiently, leading to new insights and discoveries. The findings of this study have been accepted for presentation at the International Joint Conference on Artificial Intelligence (IJCAI) 2025, underscoring the model’s significance and the broader impact of AI in advancing astronomical knowledge. FLARE represents a significant step forward in our ability to predict and understand stellar flares, offering scientists a powerful new instrument to explore the intricate dynamics of stars and their environments.