Environmental and energy performance analyses of HVAC systems in office buildings using boosted ensembled regression trees: Machine learning strategy for energy saving of air conditioning and lighting facilities

Abstract

HVAC systems consume around 40 % of a building’s energy, making them a primary target for reducing energy use and CO₂ emissions. Traditional optimization and fault detection methods may no longer meet critical efficiency and thermal comfort standards. Artificial intelligence models, supported by improved sensor technology, offer powerful solutions by leveraging large datasets. This study thoroughly examines various energy consumption prediction methods, advancing from traditional analytical models to boosted-ensembled machine learning techniques. Four different regression tree models, including K-Nearest Neighbors Regression (K-NN), Extra Trees Regressor (ETR), Voting Hybrid Regression (VHR), and Multi-Layer Perceptron Regression (MLPR), are utilized to predict energy consumption for HVAC and lighting in office buildings. The study uses real meteorological information and office building energy consumption based on two years of historical energy use data. Moreover, eight distinct statistical indicators are used to evaluate the robustness of the four models. The study also analyzes how different features, such as building location, size, and weather variables, affect each model's ability to improve long-term energy consumption forecasting. The results demonstrate that the ETR approach provides the highest prediction accuracy, followed closely by VHR and MPLR, whereas KNN shows the lowest accuracy. The coefficient of determination (R²) and the root mean square error (RMSE) for predicting HVAC energy consumption are 0.9943 and 0.4352 for ETR and 0.9943 and 0.4500 for VHR, respectively. Contradictively, KNN displays values of 0.988 for R² and 0.586 for RMSE, underscoring a clear performance hierarchy among these predictive methods. Conclusively, the ETR method emerges as a powerful optimization tool for accurately predicting the energy consumption of HVAC and lighting in office buildings.