Exploring Forecasting and Prediction Processes for Decision-Making to Promote the Photovoltaic Energy Integration into the Grid: A Mini Review

As solar energy is increasingly integrated into the electricity grid, the power system faces several technical challenges posed by the fluctuating renewable source. This paper reviews the forecast and prediction processes, which are crucial components for Grid-Connected PV power plants. According to the literature, photovoltaic predictions can be made for various time scales, ranging from very short-term forecasts to short-term, medium-term, and long-term projections. The temporal resolution depends on the specific needs and applications of the prediction, such as optimizing energy management or planning for grid integration. The spatial resolution corresponds to the sampling area of the prediction. The accuracy of a photovoltaic (PV) forecast can be measured by comparing the predicted values with the actual observed values. Several performance parameters provide quantitative measures to aid in evaluating the efficacy of the forecasting model, such as RMSE, r2, MAE, and MAPE. Drawing on existing literature, the scientific community experiences various forecast techniques, which are classified mainly into persistence methods, physical techniques, statistical techniques, and hybrid models, with a prevailing trend toward hybrid deep-learning models in last years. However, certain gaps exist, particularly related to complexity and time consumption. With increasing PV penetrations, the impact of incorrect forecasts in the grid can be larger. In some electricity markets, solar producers can face penalties when deviations between forecast and produced energy exceed a tolerance band. Numerous companies actively participate in the field of PV forecasting. Among them are Schneider Electric, Enphase Energy, SteadySun, and GreenPowerMonitor.