Optimizing the sustainable performance of public buildings: A hybrid machine learning algorithm

Addressing energy challenges in public buildings is essential for achieving sustainable development. This paper presents a hybrid intelligence prediction and optimization framework based on Building Information Modeling-Design Builder (BIM-DB) simulations. The framework incorporates a hybrid algorithm that combines Bayesian optimization (BO), categorical boosting (CatBoost), and the nondominated sorting genetic algorithm-III (NSGA-III). A dual-driven knowledge‒data framework is established to predict and optimize sustainability objectives, such as those related to building energy consumption (BEC), carbon emissions (CE), the thermal comfort value (PMV), and the glare index (GI). A public building in Wuhan city was selected to assess the practical feasibility of the proposed method. The results indicate the following: (1) Building parameters and thresholds were established through knowledge-driven methods, and a sample dataset was generated via BIM-DB simulations and Latin hypercube sampling (LHS), with the BO-CatBoost hybrid algorithm achieving fitness values above 0.9 for all the targets. (2) The BO-CatBoost-NSGA-III algorithm optimized sustainability performance, yielding reductions of 32.20 % in BEC, 48.77 % in CE, 60.69 % in PMV, and 15.45 % in GI. (3) A comparison with other methods demonstrated the high accuracy and reliability of the BO-CatBoost approach, whereas SHAP analysis identified key factors such as air conditioning settings and building envelope and orientation. Integrating knowledge-driven and data-driven approaches can significantly enhance the sustainability of public buildings, offering a novel research framework for future studies.