Accurate solar yield prediction is paramount for optimizing electricity generation and sizing photovoltaic (PV) power systems. With the growing prevalence of utility-scale PV installations, the inherently variable nature of solar yield necessitates robust prediction models. Precise forecasting enhances energy market trading, optimizes solar power system performance, supports grid stability, and facilitates efficient energy management, financial planning, investment decisions, and maintenance operations. This study introduces an approach for predicting solar yield using Artificial Neural Networks (ANNs), exploring various ANN architectures to determine the most suitable structure and comparing its performance against several machine learning models. Through the use of Spearman and Pearson Correlation feature selection techniques, we identified all input features as being essential for the accurate prediction of Global Horizontal Irradiance (GHI). The findings indicate that the highest performing ANN attains a RMSE of 25.92 W/m², a MBE of −1.55 W/m², a MAPE of 0.11%, and a R-squared of 0.9926. This result demonstrates that our ANN model significantly outperforms other models, including a highly-tuned Random Forest (RMSE: 32.83 W/m², R²: 0.988), establishing its superior performance for solar yield prediction.
For more on this article click here.