Application of hybrid machine learning algorithm in multi-objective optimization of green building energy efficiency

Green building design strives to optimize energy efficiency, emissions reduction, cost-effectiveness, and thermal comfort by accurately predicting and optimizing building performance across multiple factors. This study proposes a multiobjective prediction and optimization framework for green buildings using building information modeling-Design Builder (BIM-DB), Bayesian-random Forest (Bayesian-RF), and Non-dominated Sorting Genetic Algorithm III (NSGA-III). Firstly, BIM-DB is used for building simulation and orthogonal tests to generate data samples. Secondly, Bayesian-RF model is trained on the dataset to predict building performance. Finally, the prediction model is then used to establish the fitness function for NSGA-III optimization, which identifies the optimal solution for the multiobjective green building problem. The case study of green building design of a teaching building shows that: (1) Orthogonal building simulation experiments based on BIM-DB efficiently generate building sample datasets. (2) The Bayesian-RF method improves prediction accuracy, with MSE values below 0.08 and R2above 0.85 for all three prediction objectives. (3) The Bayesian-RF-NSGA-III optimization algorithm reduces the energy consumption of the case building by 7.68 %, carbon emissions by 6.48 %, cost by 1.77 %, and improves overall thermal comfort. The framework provides a valuable reference for setting building parameters and facilitating multiobjective optimization in green building design similar to the case buildings.