Effect of physical, environmental, and social factors on prediction of building energy consumption for public buildings based on real-world big data. (15th December 2022)

• Record Type: Journal Article
• Authors: Zhang, Yuhang; Zhang, Yi; Yi Zhang, ; Zhang, Chengxu
• Is Part Of: Energy. Volume 261:Part B(2022)
• Publication Date: 2022-12-15
• Subjects: Building energy consumption prediction -- Extra-trees -- Physical factors -- Environmental factors -- Social factors -- Big data; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2022.125286
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: In practical operations, various static and dynamic parameters affect the prediction of building energy. In this study, seven modes were constructed based on the Extra-Trees algorithm to investigate the impacts of 19 factors involving physical, environmental, and social aspects on the specific energy consumption. The analysis is supported by data obtained from 110 public buildings over one year in Shenzhen, China. The results indicate that considering more factors significantly improves the prediction accuracy. Factor importance analysis shows that building type is the most significant factor, followed by ambient temperature, month, structure, floor area, and rent price. The external wall material is the least significant factor, followed by the type of heating system and subdistrict population. At least the top 8 most important factors must be provided to maintain an accuracy above 80%. Furthermore, the sensitivity analysis indicates that the SECs of comprehensive and office buildings demonstrate opposite variation trends to floor area; for rent price and distance from central business district, the SECs of comprehensive building show inverse and positive change trends, respectively, but the SECs of office remain almost unchanged. In addition, buildings with internal insulation can lead to more SEC in such climate. Highlights: 19 factors involving physical, environmental and social category are considered. Seven Extra-Trees prediction models are built to investigate influencing factors. The importance analysis is conducted to rank all the factors. Sensitivity analysis is made to find the variation trend under eight scenarios. Recursive feature elimination is made to quantify the factor impact on prediction.