Energy consumption prediction for office buildings: Performance evaluation and application of ensemble machine learning techniques
Document Type
Article
Publication Date
5-1-2025
Abstract
Accurate forecasting and evaluation of building energy consumption are paramount for enhancing energy efficiency, reducing operational costs, and mitigating environmental impacts. Effective energy management relies on precise predictions to inform decision-making and optimize resource allocation. Although promising predictive capabilities have been demonstrated by ensemble models in this domain, their practical application is often hindered by prolonged training times and high computational demands. To address these issues, a novel ensemble modeling strategy was developed herein, incorporating the Adaptive Gradient Boosting Regression (AGBR) algorithm. The AGBR model was built with a two-layer structure and iterative residual modeling, incorporating adaptive early stopping mechanisms and gradient-regulated learning rates. These innovations improve training efficiency and predictive accuracy by enabling dynamic adjustments based on validation errors. Furthermore, Kernel Principal Component Analysis (Kernel PCA) was utilized for feature reduction within an explainable ensemble model framework, thereby facilitating accurate predictions of office building energy consumption. This methodology not only identifies the most influential feature variables but also evaluates their relative importance by revealing underlying nonlinear relationships that may be overlooked by traditional linear methods. The proposed model was validated using data from an office building in Beijing Province, achieving a remarkable 73.91 % reduction in training time and a 3.13 % improvement in predictive accuracy compared to standard Gradient Boosting models. Additionally, the stability of predictions was significantly enhanced, as evidenced by a 62.28 % reduction in Mean Absolute Error (MAE). These findings demonstrate the potential of the proposed model to enhance building energy management and optimize performance effectively.
Keywords
Building energy consumption prediction, Machine learning, Ensemble learning, Performance evaluation
Divisions
mechanical
Funders
Key Laboratory for Comprehensive Energy Saving of Cold Regions Architecture of Ministry of Education,The 13th Five-Year National Science and Technology Major Project of China [Grant No: 2016YFC0801706; 2017YFC0702202],National Natural Science Foundation of China (NSFC) [Grant No: 51578011],Hebei province international science and technology cooperation fundamental project at the North China Institute of Science and Technology [Grant No: 20594501D]
Publication Title
Journal of Building Engineering
Volume
102
Publisher
Elsevier
Publisher Location
RADARWEG 29, 1043 NX AMSTERDAM, NETHERLANDS