Data-driven estimation of building energy consumption and GHG emissions using explainable artificial intelligence

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review

32 Scopus Citations
View graph of relations

Author(s)

  • Yan Zhang
  • Bak Koon Teoh
  • Maozhi Wu
  • Jiayu Chen
  • Limao Zhang

Detail(s)

Original languageEnglish
Article number125468
Journal / PublicationEnergy
Volume262
Online published19 Sept 2022
Publication statusPublished - 1 Jan 2023

Abstract

Energy consumption prediction is an integral part of planning and controlling energy used in the building sector which accounts for 40% of the global energy consumption and a significant portion of greenhouse gas emissions. However, very few studies focused on the combined effect of building characteristics, building geometry, and urban morphology on energy performance. Such a research gap is addressed in this study by developing an explainable deep learning model. Our model uses Light Gradient Boosting Machine integrated with the SHapley Additive exPlanation algorithm, so as to provide insights into the feasibility of using machine learning-based models for energy performance prediction of buildings. With the proposed eXplainable Artificial Intelligence model, this study successfully predicts energy usage and greenhouse gas emissions of residential buildings, as well as identifies the most influential variables and evaluates their relative importance. A case study based on Seattle's data is used to verify the proposed framework, and some conclusions can be drawn: (1) Urban morphology and building geometry have significant effects on evaluating the building energy consumption and greenhouse gas emissions, as the accuracy of predicted result improve 33.46% compared with only considering building characteristics; (2) The total gross floor area and natural gas are identified as the most influential factors for energy consumption and GHG emissions, respectively; (3) The proposed model is examined to be an accurate method with the R2 of 0.8435 on average, comparing with the other approaches, such as the eXtreme Gradient Boosting, Random Forest, and Support Vector Regression. The main contributions of this research lie in that (a) a comprehensive structure integrated with building characteristics, building geometry, and urban morphology is established to forecast the energy use and greenhouse gas emissions; (b) an explainable artificial intelligence model incorporated with the SHapley Additive exPlanation algorithm into Light Gradient Boosting Machine has been proved to achieve an accurate prediction of the energy performance of residential buildings.

Research Area(s)

  • Urban morpho-blocks, Data -driven estimation, Light gradient boosting machine, Explainable AI, Building energy consumption, OFFICE BUILDINGS, URBAN DENSITY, PREDICTION, SCENARIOS, IMPACTS, GIS, ANN

Citation Format(s)

Data-driven estimation of building energy consumption and GHG emissions using explainable artificial intelligence. / Zhang, Yan; Teoh, Bak Koon; Wu, Maozhi et al.
In: Energy, Vol. 262, 125468, 01.01.2023.

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review