CECA: An intelligent large-language-model-enabled method for accounting embodied carbon in buildings

The construction sector's contribution to global carbon emissions is significant, with embodied carbon accounting for around 40 % of the sector's total emissions and even more than 50 % in the case of net-zero energy buildings. Traditional life cycle assessment methods for embodied carbon accounting are time-consuming and labor-intensive, hindering rapid evaluation of carbon footprint in buildings. This study presents a novel construction embodied carbon assessment (CECA) method. It uses large language models for intelligent semantic parsing and automatically matches material and equipment information with corresponding carbon emission factors. Based on our experiments, CECA outperforms traditional machine learning algorithms, achieving a matching accuracy of 0.8412. Validation across 18 real-world building cases demonstrates that the CECA method with Claude-3.5 achieves comparable carbon accounting accuracy to traditional life cycle assessment methods, with an average mean absolute percentage error of 12.5 %. Additionally, CECA significantly enhances computational efficiency, achieving a 216-fold improvement—approximately 60 ss for CECA compared to 3.6 h for LCA in our experiment where adopted BIM models are already available. Furthermore, the CECA method can identify carbon emission contributions of various materials and components, facilitating carbon optimization during the building design process. By advancing the application of large language models in evaluating embodied carbon in buildings, the CECA method offers a promising solution for rapid, accurate, and automated embodied carbon accounting, addressing the growing demand for efficient carbon evaluation and management in construction projects.