Combination of a Novel All Sky Imager Based Approach for High-resolution Solar Irradiance Nowcasting with Persistence and Satellite Nowcasts for Increased Accuracy

Electricity grids with high PV-penetration benefit from the consideration of intra-minute and intra-hour variabilities via nowcasts (shortest-term forecasts). In this study we present approaches for blending all sky imager (ASI), persistence and satellite based nowcasts for increased nowcasting accuracy within the 15-minute interval. Our ASI method is a novel machine-learning (ML) based model, trained on spatially distributed irradiance measurements in Freiburg, Germany. These come from an irradiance measurement network of eight stations within a radius of ~10 km around the camera position, which we also use to evaluate our forecasting results. Our ASI method exhibits a significantly lower root mean square error (RMSE) than the satellite-based method up to a lead time (LT) of 11 minutes ahead and a lower mean absolute error (MAE) throughout the entire interval. Using a LT dependent linear combination of the individual models RMSE improvement scores of 5 - 13% and MAE improvement scores of up to 6% (for LT ≥ 5 min) could be achieved over the respective optimal individual method. Further improvements in RMSE (up to 2%) and MAE (up to 7.5%) were achieved by using a Lasso model with 3rd degree polynomial features and including cloudiness, sun position and variability of clear-sky index as additional input parameters.