Benchmarking machine learning algorithms for high-fidelity power forecasting in utility-scale PV plants

Abstract

Accurate prediction of photovoltaic (PV) power is essential for effective monitoring, control, and grid integration of large-scale solar systems. This study uses high-resolution data from a 20 MW utility-scale PV plant to benchmark four advanced machine learning (ML) models: Support Vector Machine (SVM), Random Forest (RF), Gaussian Process Regression (GPR), and Artificial Neural Network (ANN). Model performance was evaluated using R², RMSE, and MAE. Additionally, a thorough analysis of the three-dimensional (3D) predicted power surfaces was conducted. The findings indicate that GPR and ANN outperform SVM and RF, achieving near-perfect accuracy (R² = 1.0000) with minimal error. These methods yield smooth, physically consistent surfaces that accurately represent PV system behaviour. These findings establish GPR and ANN as the most reliable machine learning (ML) approaches for high-fidelity photovoltaic (PV) power forecasting and energy management applications.