Clouds moving in front or away from the sun are the leading cause of irradiance variability. These variations have a repercussion on the electricity production of photovoltaic systems. Predicting such changes is essential for proper control of these systems and for maintaining grid stability. Images from the sky have proven to help with short-term solar irradiance forecasting, especially when combined with artificial intelligence. Nevertheless, these models tend to smooth the irradiance fluctuations. We propose a forecasting model to predict the clear-sky index in a forecast horizon of 20 min with a 1-minute resolution. Our model, based on a classifier to determine the sky conditions and, on an optical flow, applies an artificial intelligence model explicitly trained on each class of sky conditions. This strategy has an equivalent performance to an unclassified model and a forecast skill between 5 and 20% with respect to the smart persistence model for most classes of sky conditions while requiring considerably less training data. Although our model reduces the overall predicting error, it still has difficulties predicting irradiance changes and mainly overcast days. Our classifying strategy can be applied to other models targeting different objectives to predict sudden changes in either irradiance or power related to photovoltaic systems.

Short-term solar forecasting is essential for the large-scale application of solar energy and is necessary for the operation of power plants, energy trading, and grid balancing. The main cause of uncertainty in solar forecasting is cloud movement, which can be observed by All-Sky Images. The spatial layout and temporal dynamics of clouds cannot be extracted by conventional cloud modelling approaches utilising image analysis techniques, leading to inaccurate predictions of the interaction with solar radiation. In this study an categorization method is made based on sky conditions and from that five classes are made. The goal of this classification method is to improve the 21-minute irradiance predictions which are made with a deep learning model. The outcome of the 21-minute deep learning model will get compared with the Persistence, Smart Persistence and ARIMA model. This classification method and irradiance prediction are applied for location Folsom, California and Delft, Netherlands. The irradiance predictions based on sky classification showed an improvement of 8.6% for location Delft and 29.3% for location Folsom when the same dataset sizes are used. The irradiance predictions were also done for various dataset sizes and compared per location. Finally, it provides a way forward to improve the classification method and deep learning models to anticipate short-term irradiance in the future.