Machine learning based evaluation of window parameters on building energy performance and occupant thermal comfort under climate change

This study aims to evaluate the effect of climate change on energy consumption and thermal comfort of an existing educational building and to use machine learning techniques for building performance improvement through passive retrofit scenarios based on window parameters. Theory and Methods: The research proposes a four-step method based on performance analysis through building simulations: (i) the generation and analyses of climate change modified future weather files, (ii) climate change impact analysis on an existing building, (iii) comparative analysis of retrofit scenarios, and (iv) the analyses prediction models. The proposed method is implemented in a secondary school building in Ankara, Turkey. Results: Simulation results show that by 2050, the annual heating (QI) will decline by 33%, whereas the annual cooling (QS) will rise by 100%. The June indoor overheating degree (IADH) values will increase from 0.74 degrees C to 2.32 degrees C, showing a drastic change of 213%, implying a significant increase in thermal discomfort. Correlation analysis indicates that both IADH and QS have a strong positive linear relationship with SHGC, whereas the QI has a strong positive linear relationship with Upencere. Furthermore, the QI-SHGC relationship is moderate negative. Developed prediction models for energy consumption and thermal comfort have a high predictive capacity with a minimum R2 of 0.927 and, on average, 2% deviation from their actual values based on RMSE. Random forest feature importances for the prediction models trained to predict QS and IADH agree on the importance of the SHGC. For QI, Upencere has a slightly higher significance than the SHGC. Conclusion: Building energy consumption and occupant thermal comfort are significantly affected by climate change. The fact that the majority of the school building occupants are composed of young children not only distinguishes school buildings from other building types but also makes these buildings more vulnerable to heat stress. The results of this study emphasize the necessity of retrofit measurements against climate change. The random forest feature importances indicate that the window SHGC value is the most critical window parameter of the performance-based improvement scenarios among the tested variables. The steps suggested and followed in this study can be applied to the other building typologies with different retrofit scenarios. The scope of this research can be expanded with varying building parameters and technologies.