The rising global demand for air conditioning systems, driven by increasing temperatures and urbanization, has led to higher energy consumption and greenhouse gas emissions. HVAC systems, particularly AC, account for nearly half of building energy use, highlighting the need for efficient cooling solutions. Passive cooling, especially radiative cooling, offers potential to reduce cooling loads and improve energy efficiency. However, most studies focus on idealized conditions, neglecting the real-world variability of indoor and outdoor environments. This study proposes a novel machine learning-based ensemble stacking model to predict ventilation rates in passive cooling buildings, addressing the challenges of black-box modeling. The model's performance is improved across key metrics such as R2, RMSE, and MAE. For the first time, uncertainty and sensitivity analysis is applied to assess the impact of indoor and outdoor conditions on ventilation rates. Sensitivity analysis shows that the reference model's ventilation rate highly depends on inlet air temperature, internal temperatures at 0.1 and 0.2m, and internal wall heat flux, with optimization of these parameters having a significant impact on building performance. In contrast, the test building relies on fewer parameters, with external temperature, outlet air temperature, and net roof radiation being notable factors; as ambient temperature increases, so does the ventilation rate. The analysis reveals that uncertainties have minimal impact in the reference building, while the test building demonstrates greater sensitivity during warmer months, emphasizing the importance of accounting for seasonal variations. This research underscores the significance of optimizing key features to enhance natural cooling and ventilation, contributing to sustainable climate control solutions and providing an interpretable, robust model for predicting ventilation rates in energy-efficient buildings.
Read full abstract