Physics-informed large models for wind and solar forecasting: A carbon-aware framework for decarbonized power systems

The large-scale integration of wind and solar power is essential for power-sector decarbonization, but their variability creates persistent challenges for forecasting, reserve allocation, storage scheduling, curtailment, and emission-aware operation. This review synthesizes recent advances in statistical, deep-learning, transformer, foundation-model, probabilistic, and physics-informed renewable forecasting, with emphasis on physical consistency, uncertainty, computational requirements, real-time applicability, and system-level operational value. We propose the Physics-Informed Large Model Carbon-Aware (PILM-CA) framework, a conceptual architecture linking physical renewable dynamics, physics-informed large-model forecasting, grid-operational decision support, and carbon-aware optimization. Unlike accuracy-centric forecasting frameworks, PILM-CA explicitly represents the pathway through which forecast uncertainty can affect commitment, reserves, storage, fossil backup, curtailment, and consequent carbon intensity. Published evidence indicates that these effects are system-dependent rather than universal; therefore, the framework distinguishes evidence-supported components from conceptual cross-layer relationships requiring operational validation. Probabilistic forecasting metrics, extreme-event resilience, digital-twin deployment, technology readiness, and climate adaptation are also critically examined. The review identifies research priorities for scalable, physically coherent, uncertainty-aware forecasting that supports both decarbonization and resilience in high-renewable power systems.