人工智能（AI）正成为人类破解宇宙奥秘的"智慧伙伴"。日前，在瑞士日内瓦举办的2025年人工智 能向善全球峰会上，之江实验室联合中国科学院国家天文台研发的OneAstronomy天文领域大模型，成 功入选"人工智能向善创新实践案例集"。 在"AI+天文"的探索征程中，中国既收获了亮眼成果，也直面发展瓶颈，更在谋划着数据、算法与 实验协同的未来路径。中国科学院国家天文台研究员罗阿理近日接受科技日报记者采访时说，要推动 AI深度赋能天文研究，需进一步加强国际合作，共建天文大模型社区和遍布全球的观测设施，共享天 文数据和基础设施，共同发展基于AI的天文教育和公众天文学。 天文研究迎来范式革命 现代天文观测已进入"大数据时代"，传统研究模式面临严峻挑战。大型巡天望远镜每年产生的拍字 节（PB）级数据，若依靠人工分析，即便耗费数年也难以完成筛选与解读。AI为天文研究带来了全方 位"智能革命"，"智能钥匙"正在打开宇宙探秘新空间。 "AI对天文研究的赋能，首先体现在高效处理海量数据方面。"罗阿理说，面对复杂的宇宙数据，AI 算法如同不知疲倦的"筛选员"，能快速完成搜索、分类与异常检测等基础工作。无论是从繁杂数据中精 准识 ...

Core Viewpoint - The FLARE model represents a significant advancement in predicting stellar flares, showcasing the potential of AI in astronomical research [2][3][4]. Group 1: Model Development - The FLARE model was developed by a collaborative team from the Purple East Taichu and the National Astronomical Observatories of China [2]. - It utilizes a unique Soft Prompt Module and Residual Record Fusion Module to enhance the extraction of light curve features, improving the accuracy of flare predictions [14][17]. - The model architecture involves decomposing light curves into trend and residual components, applying moving average methods to mitigate data loss, and integrating historical flare records to bolster robustness [15][17]. Group 2: Stellar Flares and Prediction Challenges - Stellar flares are rapid releases of magnetic energy in a star's atmosphere, crucial for understanding stellar structure, evolution, and the search for habitable exoplanets [7]. - The limited number of observed flare samples has hindered comprehensive research, making accurate prediction of stellar flare timing a critical task [8][9]. - Unlike solar flares, predicting stellar flares primarily relies on light curves, which often suffer from data gaps and significant variability across different stars [10][12]. Group 3: Model Performance - The FLARE model was tested using high-precision light curve data from 7,160 stars, demonstrating superior performance compared to various baseline models, including MLPs, RNNs, CNNs, GNNs, and Transformers [18][20]. - It achieved an accuracy of over 70%, significantly outperforming other models across multiple evaluation metrics such as F1 score, recall, and precision [20]. - The model's adaptability allows it to accurately predict flare events based on varying light curve patterns, even for the same star under different conditions [21][22]. Group 4: Future Implications - As research progresses, the FLARE model is expected to play a larger role in astronomical studies, aiding scientists in exploring more cosmic mysteries [23].