Enhancing Solar Power Forecasting Accuracy Through Machine Learning Architectures
Year: 2026
Type: conference-paper
Abstract: Accurate photovoltaic (PV) power forecasting is essential for ensuring stable power system operation and efficient energy management due to the intermittent nature of solar energy generation. The research investigates which machine learning models perform best for multiple future time periods when predicting PV power output in Dakhla which exists in a coastal Moroccan region that possesses semi-arid weather patterns and substantial solar energy resources. The dataset contains hourly data for the entire year 2022 which produces 8760 total observations. The input variables include solar irradiance components, solar height, ambient temperature and wind speed, while the target variable is PV power output. The researchers developed a comparative forecasting framework which utilized four machine learning models: Extreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP), Decision Tree (DT), and Support Vector Regression (SVR). The models were tested to perform forecasting for three different time intervals which included 1-hour, 1-day, and 1-month periods. The results indicate that forecasting accuracy strongly depends on the prediction horizon. MLP and SVR achieved the best performance for 1-hour forecasting with RMSE values around 20 W/m2and R2values above 0.96, while SVR and XGBoost provided the most reliable day-ahead predictions with RMSE values below 23 W/m2and R2values close to 0.97. The models exhibited lower accuracy during the monthly forecast period because of increased prediction uncertainty yet SVR and DT maintained stable trend estimation while MLP better preserved long-term variability. The study demonstrates the importance of horizon-aware model selection for improving the reliability and accuracy of solar power forecasting systems.
Institution: Université Ibn-Tofail
Language: English
Field: Energy
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204145873