ABSTRACT
This paper proposes a UQ framework for the Baguan-solar model by leveraging ensemble sampling from weather foundation models, generating probabilistic forecasts and deriving calibrated prediction intervals.
PAPER · PDF
Loading PDF...
Key findings
Proposes Ensemble Baguan-solar for probabilistic solar irradiance forecasting.
Combines initial condition perturbations, stochastic neural network sampling, and multi-model aggregation.
Evaluates the method using CRPS, Brier score, and reliability diagrams against deterministic baselines.
Expected outcomes include a 15-25% improvement in probabilistic forecast skill.
Limitations & open questions
Computational constraints for operational deployment need to be addressed.