人工智能在空间天气建模中的应用综述

In recent years, the deep integration of Artificial Intelligence (AI) technologies into space weather and space exploration is accelerating a paradigm shift in space science, from the traditional “physics–numerical simulation” model toward a new “physics-guided data-intelligent modeling” framework. This paper comprehensively synthesizes research progress across various subfields of space science and systematically reviews AI-driven modeling and prediction applications in key processes such as solar physics, coronal dynamics, solar–terrestrial coupling, geomagnetic disturbances, magnetic storms and substorms, ionospheric responses, magnetic reconnection and turbulence evolution, and planetary space environments. Special attention is given to the modeling capabilities of mainstream AI algorithms, including convolutional neural networks (CNNs), deep learning, Transformers, and physics-informed neural networks (PINNs), in capturing multi-scale, complex physical phenomena. Their strengths and limitations in event identification, time-series forecasting, physical structure reconstruction, and causal inference are discussed in detail. The paper further analyzes common challenges in current space weather AI models, including poor interpretability, imbalanced datasets, and limited generalization capability, and emphasizes the foundational importance of building standardized, “AI-ready” databases. Using examples such as NASA ′s SDO mission, the SPASE metadata model, and China ′s Meridian Project (CMP), the paper explores how multi-source, multi-scale observational data and physical simulation frameworks provide critical support for AI model training and validation. It advocates leveraging the Meridian Project as a strategic platform to build an intelligent infrastructure for China′s space science of “mission-driven, data-grounded, and AI-powered.”