Quality and economic value of operational photovoltaic power forecasts in the day-ahead market

In modern power systems with renewables, solar forecasting precedes scheduling and bidding. Whereas accuracy has always been the chief pursuit in solar forecasting, this study aims to understand how forecast value is linked to forecast quality by conducting operational solar forecasting within the Hungarian day-ahead market, an integrated part of the European market. Three numerical weather prediction models (a regional, a global deterministic, and a global ensemble), three solar power curves (physical model chain, machine learning, and hybrid) are combined with and without a final post-processing step, optimized for two directives (mean absolute and mean squared error minimization), leading to 36 distinct forecasting workflows. The quality assessment performed for 12 photovoltaic power plants reveals that the forecast accuracy is most affected by numerical weather prediction model selection, followed by the irradiance-to-power conversion method, and lastly, post-processing, provided that the models in all steps are properly optimized. The forecasting workflow affects the economic value of the forecasts, but maybe surprisingly, the plant siting and design (i.e., location and panel orientation) exert an impact on the cost of forecast errors no smaller than the forecasting itself. The joint scheduling of multiple plants does not reduce the imbalance costs in the single-pricing imbalance settlement system, which contrasts with the forecast error reduction due to geographical smoothing.