The National Astronomical Observatory of the Chinese Academy of Sciences and Alibaba Cloud have released the world's first large-scale solar AI model "Golden Crow". Based on Alibaba's Tongyi Qianwen series of open source models, the accuracy rate in M5-class solar flare prediction is currently over 91%, which is also the highest level of solar flare prediction at this level.Solar flares are violent energy release events that occur on the surface of the sun, radiating a large number of charged high-energy particles, affecting the earth's electromagnetic environment. In severe cases, they can cause widespread power outages, threatening the safety of orbiting satellites and space stations.

Solar flares are usually divided into five levels: A, B, C, M, and

Since 2000, solar observation data has increased exponentially, but manual labor alone is no longer able to process the massive observation data. AI has provided new tools for solar forecasting.

Compared with traditional machine learning, large models are better at processing multi-modal data such as images, and have the ability to understand and reason about complex physical relationships, making them very suitable for solar forecasting.

Based on Alibaba Cloud Qwen-VL and other models, the "Golden Crow" sun model is based onMore than 900,000 solar satellite images as samples, the fine-tuning training is completed.

By inputting the solar physical parameters of the previous period and the corresponding observation images, "Golden Crow" can predict the outbreak of flares in the next 24 hours.

At the same time, "Golden Crow" can infer the physical parameters of the sun for the next period, and generate simulated solar images for the next period by calling the diffusion model.

Lin Jiaben, director of the Huairou Observation Base of the National Astronomical Observatory, pointed out: "The Ali Tongyi large model has excellent performance in multi-language and image recognition capabilities such as Chinese and English, and is very suitable for solar forecasting. As a scientific research institution, compared with training large models from scratch, developing scientific models based on existing basic models is a more efficient and feasible choice."

In the future, the "Golden Crow" large solar model will also be used to study difficult problems such as the origin of the solar magnetic field.

It is worth mentioning that the "Star Language" large model of the National Astronomical Observatory of the Chinese Academy of Sciences, the world's first large-scale water energy and grain model "Luoshu" of the Tibetan Plateau Institute of the Chinese Academy of Sciences, and the large multi-modal coral reef model "Yaohua" of the South China Sea Institute of the Chinese Academy of Sciences are all based on the Ali Tongyi Qianwen series of models.