Relationship between built form and energy performance of office buildings in a severe cold Chinese region

详细信息    查看全文

• 作者：Wei Tian ; Song Yang ; Jian Zuo ; ZhanYong Li ; YunLiang Liu
• 关键词：built form ; energy performance ; simulation model ; sensitivity analysis ; machine learning
• 刊名：Building Simulation
• 出版年：2017
• 出版时间：February 2017
• 年：2017
• 卷：10
• 期：1
• 页码：11-24
• 全文大小：2,828 KB
• 刊物类别：Engineering
• 刊物主题：Building Construction, HVAC and Refrigeration
  Engineering Thermodynamics and Transport Phenomena
  Atmospheric Protection, Air Quality Control and Air Pollution
  Environmental Computing and Modeling
  Chinese Library of Science
  
• 出版者：Tsinghua University Press, co-published with Springer-Verlag GmbH
• ISSN：1996-8744
• 卷排序：10

文摘

It is well recognized that building form has significant influences on energy performance in buildings, especially in the cold climate. It is imperative to understand the relationship between built forms and energy use in order to provide guidance in early project stage such as preliminary design. Therefore, this study focuses on two aspects to understand characteristics of energy use due to the change of parameters related to building form. The first aspect is to apply new metamodel global sensitivity analysis to determine key factors influencing energy use and the second aspect is to develop reliable fast-computing statistical models using state-of-art machine learning methods. An office building, located in Harbin, China, is chosen as a case study using EnergyPlus simulation program. The results indicate that non-linear relationships exist between input variables and energy use for both heating and electricity use. For heating energy, two factors (floor numbers and building scale) show a non-linear yet monotonic trend. For electricity use intensity, building scale is the only significant factor that has non-linear effects. It is also found that the ranking results of critical factors to both electricity use and heating energy per floor area vary significantly between small and large scale buildings. Neural network model performs better than other machine-learning methods, including ordinary linear model, MARS (multivariate adaptive regression splines), bagging MARS, support vector machine, random forest, and Gaussian process.