A hybrid machine learning model to optimize thermal comfort and carbon emissions of large-space public buildings

The use of the minimum energy to maintain the indoor thermal comfort of the large-space public building is always a challenging task due to the complex outdoor environment and indoor requirements. The lack of monitoring data and effective approaches limits the understanding of building thermal and energetic performance. This paper thus proposes a hybrid machine learning model based on factor generators and an optimization approach to address this research topic, aiming to provide the essential guide for future retrofit and design of large-space public buildings. The four machine learning (ML)-based factor generators are compared using the one-year monitoring data of building facility and indoor thermal management, where the high-performance multilayer perceptron neural networks (MLPNN) model is chosen as the data-driven method to generate the input data as the parent or intermediate populations in the GA optimization algorithm. Such a hybrid machine learning model can solve the multi-objective functions of thermal comfort and carbon emissions. The optimization results demonstrate that this model can achieve a maximum 29% improvement for thermal comfort and a reduction of 386.9 kg CO2 (11.06%) for carbon emissions in comparisons with the human-based management. Moreover, such hybrid machine learning model exhibits tolerance for moderate deficit in one objective. Therefore, the optimal thermal comfort and carbon emissions of large-space public buildings are achieved and thus contributing to the carbon neutrality in the building sector.