Answers are grounded in this paper and cite their sources.
Accurate solar energy prediction remains a major challenge due to the nonlinear behaviour of photovoltaic systems and the variability of atmospheric conditions. In this paper, a systematic and experimentally verified comparison methodology for the PV power forecast using three popular machine learning (ML) algorithms, Artificial Neural Networks (ANN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), is presented. The novelty of this work lies in combining high-resolution real experimental data collected under semi-arid climatic conditions with a dual evaluation strategy that assesses model performance both at a global level and under four representative weather regimes. This approach enables a deeper analysis of model robustness, generalisation capability and sensitivity to environmental variability that is frequently overlooked in the existing studies. The results demonstrate that RF consistently achieves the highest predictive performance, with R² = 0.9997 and RMSE = 0.7757, exhibiting superior accuracy, minimal bias and strong stability across all weather conditions. XGBoost shows competitive performance but slightly reduced robustness under highly dynamic conditions, while ANN presents higher prediction errors and greater sensitivity to nonlinear fluctuations. These findings demonstrate that ensemble learning provides an effective, robust and computationally efficient solution for real-world PV forecasting, making it highly suitable for solar energy management applications.
No takes yet. Share an insight, caveat, or question.
Elwardi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: