MACHINE LEARNING ALGORITHM COMPARISONS IN THE FIELD OF LOW-ENERGY BUILDING DESIGN AND OPERATION
Date
2023-08-14
Authors
Birdsell, Blair
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This work spans multiple disciplines to focus on innovation in data-driven decision-making in
building performance, utilizing machine learning algorithms. It provides performance benchmarks
and analysis that can guide industry best-practice in building design and operations. Specific
outcomes from this work have broad application such as the optimization of building energy
efficiency, bettering occupant comfort, potential energy saving retrofits, and improving overall
building performance. Data-driven decisions are made by evaluating and comparing data from
various sources, including building simulations, historical records, and sensor measurements.
Through the application of machine learning tools, this data is transformed into a foundation for
effective decision-making by building designers and operators or incorporated into intelligent
building management systems.
Gradient Boosted Trees when applied to building energy performance demonstrated robust
prediction characteristics across different data types and problem configurations as well as being
highly accurate and computationally efficient. In predicting building performance time series, the
research documents neural network architectures containing convolutional layers as having the
ability to forecast short-term high-frequency variations in airflow and predict monthly minimal and
maximum airflow values with high accuracy. These results support their inclusion and application
in building management systems for high-performance smart buildings.
Description
Keywords
Machine learning, Building simulation