Short-Term Solar Irradiance and Power Forecasting in Uncertain Photovoltaic Systems

Renewable energy (RE), while essential for the energy transition, poses significant challenges due to its variability and reliance on environmental factors , such as sunlight availability for Grid-Connected Photovoltaic (GCPV) systems. This intermittent nature complicates maintaining a stable balance between energy production and demand. To address these challenges, Energy Management (EM) plays a vital role in enabling optimal planning and resource allocation. In this context, forecasting becomes crucial for predicting production fluctuations and supporting proactive decision-making. Short-term forecasting, in particular, facilitates quick anticipation of variations and real-time adaptation of management strategies. This study proposes a Feedforward Neural Network (FFNN) model for short-term forecasting in renewable energy systems, effectively addressing variability challenges. By incorporating prediction intervals, the model ensures accurate predictions with quantified uncertainty. The model demonstrates high performance, with irradiance Mean Absolute Error (MAE) ranging between 0.3115 and 0.3506, Root Mean Squared Error (RMSE) between 0.4061 and 0.4802, and R 2 of 1.00. For power, the MAE ranges from 7.0136 to 7.8298, RMSE from 8.2715 to 9.2084, and R 2 of 0.99, underscoring its reliability for energy management applications.