A machine learning-based surrogate model to approximate building carbon footprint in early-stage architectural design

Recent advancements have made “surrogate models” an emerging option for building performance assessments and whole life carbon assessment; these are machine-learning models fitted to a sample of results from a parametric simulation, such that they act as a fast but approximate replacement for the original detailed simulation. Many articles have developed surrogate models to assess energy performance; however, only a limited number of studies have explored surrogate modelling for whole-life carbon assessment. This gap is mainly due to the poor availability of life cycle analysis tools and datasets. In this paper, we generate simulation results from a parametric LCA model and propose a methodology to use this synthetically generated database to develop a machine learning-based prediction model of both operational and embodied carbon. We test the model using two case studies. The results show that the model achieves high prediction performance using the minimal inputs available during early-design phases. The results indicate that such techniques can be used by building designers with limited LCA expertise to instantly estimate building sustainability performance and to select design options with reduced emissions.