Climate change is one of the biggest concerns of our time. Governments all over the world are formulating policies to reduce carbon emissions. For instance, China has promised carbon neutrality by 2060. Buildings account for around half of the country’s emissions and total energy consumption. Moreover, their impact on human lives has increased following COVID-19. Therefore, building performance optimization (BPO) is crucial to reduce their carbon footprint. BPO can be achieved by assessing the overall environmental impact of buildings throughout their life cycle, resulting from material production, construction, operation, and demolition. However, there is a lack of research from this perspective.
Recently, a team of researchers, led by Dr. Yaw-Shyan Tsay of National Cheng Kung University (NCKU), Taiwan, has proposed a strategy that combines machine learning models with new optimization algorithms to simulate the whole life cycle performance of the building. Their work was made available online on 10 September 2022 and will be published in Volume 262, Part A of the Energy journal on 01 January 2023. “This work introduces artificial intelligence into architecture, representing an emerging trend of intelligent building construction. It can directly predict life cycle carbon emissions (LCCE), life cycle costs (LCC), and indoor discomfort hours (IDH) at the design stage, producing a scheme for best building performance,” explains Dr. Tsay.
The researchers first identified 28 features as input parameters for an office in the Qingdao international academician park. Next, they sampled the parameters using five different Monte Carlo methods and prepared the data set. They applied an ensemble learning model (ELM) to make predictions. The model was trained using 70% of the data set and the remainder was used as test data. After picking out the sampling method with the best prediction performance and shortest search time, it was used to select important parameters for BPO via sensitivity analysis. Among these parameters, air infiltration rate, wall thermal transmittance, and air conditioning setpoints had the most influence on LCCE, LCC, and IDH, respectively. Further, the randomized search was found to be optimal for hyperparameter optimization. The corresponding coefficient of determination, R2, was 0.980.
Having established the feasibility of the ELM, the team next applied a two-archive evolutionary algorithm for constrained multi-objective optimization (C-TAEA) to BPO to propose various schemes. Compared with the base case, the single best solution schemes reduced LCCE, LCC, and IDH by 54.6%, 18.7%, and 64.5%, respectively. Further, the best equilibrium solution reduced LCCE, LCC, and IDH by 34.7% (1946.3 kg carbon dioxide/m2), 13.9% (2112.1 Yuan/m2), and 26.6% (2082.1 hours), respectively, surpassing the performance of existing methods. It also solved the problem of different optimization ranges of other objectives.
Dr. Tsay discusses the future implications of the research. “The present work provides an intelligent and efficient BPO strategy. It can automatically generate building schemes and improve building performance quickly by evaluating the best design approaches from the life cycle perspective. The proposed techniques will be convenient for architects and engineers to work with and will help improve the quality of life of building residents,” he concludes.
Hopefully, future research on this front involving different types of buildings will lead to the optimization of more building parameters, enabling a more comprehensive implementation of their strategy.