Hybrid Physics-Informed Machine Learning Models for Plasma Systems

Hybrid approaches of physics-informed machine learning (PIML) are emerging as a promising route to accelerate and enhance modelling of complex plasma phenomena by combining data-driven learning with governing equations, conservation laws and physical constraints. This review outlines representative plasma models and constraints, prominent data sources and benchmark problems, and leading hybrids including physics-informed neural networks, operator learning, solver-embedded learned closures, reduced-order modeling and probabilistic approaches. We offer recommendations regarding the training, validation and uncertainty quantification of such models, and we review representative applications in fusion, space and industrial plasmas. Finally, we outline future research directions for efficient, reliable hybrid plasma models to address open challenges in generalization across operating regimes, long-horizon stability, limited observability and trustworthiness of deployment.

Hali tarjima qilinmagan