Abstract
This thesis presents a novel methodology to enhance solar radiation (SR) prediction with an hourly time step and optimize building energy systems, focusing on the unique challenges of mountainous environments. A hybrid Random Forest-Symbolic Regression (RF-SR) model was developed, combining the adaptability of Random Forest in handling high-dimensional, nonlinear data with the interpretability of Symbolic Regression, which reveals key relationships between solar geometry, atmospheric parameters, and irradiance levels. The model was validated using datasets from Bolzano, Italy, and Eindhoven, the Netherlands, representing mountainous and flat terrains, respectively. The RF-SR model achieved a mean R² of 0.96 and an RMSE of 35 W/m² across diverse climatic conditions. It outperformed conventional machine learning models like XGBoost and CatBoost by 538% in accuracy and reduced prediction errors by 20% compared to traditional machine learning models when trained on constrained datasets. A key application of the RF-SR model was its integration into a metamodel for optimizing photovoltaic (PV) systems and HVAC configurations in buildings. The metamodel identified optimal PV configurations that reduced grid electricity consumption by 28% and maximized PV utilization by 22%. In Bolzano, the metamodel achieved HVAC energy savings of 18%, demonstrating the impact of accurate SR predictions on building energy efficiency. The research highlights the RF-SR model's ability to adapt to data-scarce environments, reduce computational demands, and provide interpretable insights critical for energy optimization. Its success in improving prediction accuracy and system performance positions it as a versatile tool for renewable energy applications. Future research will extend the Random Forest-Symbolic Regression model to other renewable energy systems, integrate it with dynamic tools like EnergyPlus for real-time optimization, and enhance computational efficiency using parallel processing and cloud-based solutions. Efforts will also focus on incorporating uncertainty quantification, leveraging Internet of Things (IoT) technologies, and enabling real-time applications in smart grids and urban energy systems to further its role in energy optimization and solar irradiance prediction across diverse environments.