Key points are not available for this paper at this time.
This paper presents a comparative analysis of point-wise and cumulative forecasting strategies for photovoltaic energy production over a 3-day prediction horizon. The study examines whether models that achieve high predictive accuracy at individual time steps also preserve comparable performance when forecasts are temporally aggregated. This distinction is particularly relevant for operational energy planning and decision-making processes. The results indicate that models exhibiting strong performance in short-term, point-wise prediction tasks, can nonetheless yield substantial errors in cumulative energy estimates. This discrepancy exposes an important limitation of current forecasting methodologies and raises concerns regarding their suitability for medium- to long-term energy assessment. To mitigate this divergence, the paper introduces a composite loss function specifically designed to improve forecasting performance for planning-oriented applications. The analysis encompasses the full forecasting chain, from meteorological variable prediction to photovoltaic energy yield estimation. To the best of the authors’ knowledge, this is the first work to rigorously demonstrate the systematic divergence between point-wise and cumulative forecasting accuracy in photovoltaic systems and to propose a loss function explicitly constructed to jointly optimize these two objectives. The proposed framework integrates a physics-based mesoscale numerical weather prediction model with a data-driven refinement module based on Gated Recurrent Units, thereby enhancing forecast accuracy while retaining model transparency and improving explainability. Experimental results show a 73.2% reduction in cumulative forecast error relative to the baseline physical model, as well as a 4.7% increase in the Coefficient of Determination for solar irradiance predictions. These improvements enable robust and reliable forecasting across multiple temporal scales, supporting safer grid integration, more effective storage operation and evidence-based sustainable energy planning.
Aracne et al. (Sat,) studied this question.