Deriving multivariate solar generation forecast updates—the case of imbalanced data

Accurate forecasting of solar photovoltaic (PV) generation is critical for integrating renewable energy into power systems. This paper presents a multivariate probabilistic forecasting model that addresses the unique challenges posed by the temporal variability of solar power and the structural data imbalances resulting from day and night-time periods. Unlike conventional point forecasts, our approach models forecast updates—the sequential revisions of forecasts for a fixed delivery time—thus capturing the stochastic evolution of expectations. By modeling the interdependencies of these updates, we generate consistent and realistic forecast trajectories, which are especially valuable for intraday market operations and system balancing. The methodology is applied to a case study in France, demonstrating effectiveness across different spatial granularities and forecast horizons. The model uses advanced data handling methods combined with copula models, resulting in improved Energy Scores and Variogram-based Scores. These improvements underscore the importance of addressing imbalanced data and utilizing multivariate models with repeated updates to enhance solar forecasting accuracy. This work contributes to advancing forecasting techniques essential for integrating renewable energy into power grids, supporting the global transition to a sustainable energy future.