An innovative method to predict the thermal parameters of construction assemblies for urban building energy models. (October 2022)

• Record Type: Journal Article
• Authors: Wang, Chao; Ferrando, Martina; Causone, Francesco; Jin, Xing; Zhou, Xin; Shi, Xing
• Is Part Of: Building and environment. Volume 224(2022)
• Publication Date: 2022-10
• Subjects: UBEM -- Urban energy simulation -- Non-archetype approaches -- Construction assemblies -- Thermal parameters -- Machine learning; Buildings -- Environmental engineering -- Periodicals; Building -- Research -- Periodicals; Constructions -- Technique de l'environnement -- Périodiques; Electronic journals; 696
• DOI: 10.1016/j.buildenv.2022.109541
• Languages: English
• ISSNs: 0360-1323

Abstract

Abstract: Thermal parameters of construction assemblies, e.g., the U-values of roofs, walls, floors, ground floors, windows and the solar heat gain coefficient (SHGC) of windows, are significant inputs for urban building energy models (UBEMs). However, estimating these values at the urban scale is difficult. A common practice to handle this issue is the use of archetypes. Additionally, there are three approaches in building-level studies, namely, 1) estimation based on technical documents, 2) in-situ measurements, and 3) prediction by machine learning. However, the lack of documentation or long-term testing of physical parameters restricts their applications at the urban scale. This paper presents a non-archetype approach that employs two learning algorithms, i.e., k -means and random forest classification (RFC), to predict thermal parameters of construction assemblies, with several urban & building (UB) factors selected as the inputs. The steps involve: 1) partitioning the thermal parameters in the dataset into k clusters using k -means, 2) assigning a Cluster_ID to the thermal parameters in cluster j and recording its centroid μ j, 3) training the RFC, with UB factors as inputs and the Cluster_ID as outputs, 4) predicting the Cluster_ID of thermal parameters of investigated buildings via the trained model, 5) using corresponding centroids as their final thermal parameters, based on the predicted Cluster_ID. As a pilot study, the developed approach has an acceptable result, with the R 2 greater than 0.6 and even 0.8. In addition, the study also introduces approaches for acquiring UB factors at the urban scale and demonstrates a case study in Nanjing. Highlights: The relationship between thermal parameters and urban & building factors is analyzed. Machine learning is employed to objectively reflect the relationship. The algorithms and methods for acquiring urban & building factors are introduced. The approach is in a non-archetype way that can predict the values for each building.