Climate change impacts on educational building energy demand in Iran: A machine learning approach

Abstract

Abstract

This paper quantifies how future climate change may reshape heating and cooling energy demand in Iranian educational buildings using an integrated simulation–machine-learning workflow. A three-story reference lower-secondary school is located in five cities (Tabriz, Tehran, Yazd, Rasht, and Bandar Abbas) representing cold, semi-arid, hot-dry, humid Caspian, and hot-humid coastal climates. Bias-corrected and statistically downscaled climate projections derived from CMIP6 General Circulation Models (GCMs) under SSP2-4.5 and SSP5-8.5 emission scenarios are converted into historical and future Typical Meteorological Year (TMY) weather files using the Sandia method and used as inputs to dynamic building energy simulations. Daily heating and cooling loads derived from hourly building energy simulations are used to train and test seven machine learning models: multiple linear regression, random forest, gradient boosting, support vector regression, XGBoost, LightGBM, and artificial neural networks. Model performance metrics are reported at the daily prediction level, while scenario analysis results are expressed as annual aggregated energy demand to facilitate comparison across cities and climate periods. Gradient boosting and random forests achieve the highest accuracy, typically reducing RMSE by 40–50% relative to linear regression across all five cities and both heating and cooling demand types, and achieving R 2 values of 0.90 or higher. Scenario analysis indicates that annual heating demand decreases by roughly 20–35% across the studied climates, while annual cooling demand increases by about 40–80%. Hot-dry and hot-humid cities show the largest changes, with peak cooling loads 40–60% higher; several cases shift toward cooling-dominated operation. Feature-importance and partial-dependence analyses identify outdoor air temperature and global horizontal irradiance as the dominant cooling drivers, with rapid increases above approximately 26°C and 400 W/m 2 . The results support climate-resilient design and retrofit strategies for Iranian schools, emphasizing envelope upgrades, solar control, and high-efficiency cooling in warm regions while maintaining winter performance in cold climates.