A review of physics-informed machine learning for building energy modeling

Building energy modeling (BEM) refers to computational modeling of building energy use and indoor dynamics. As a critical component in sustainable and resilient building development, BEM is fundamental to support a diverse spectrum of applications, including but not limited to sustainable building d...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Applied energy Jg. 381; S. 125169
Hauptverfasser: Ma, Zhihao, Jiang, Gang, Hu, Yuqing, Chen, Jianli
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 01.03.2025
Schlagworte:
algorithms
artificial intelligence
bioinformatics
building controls
Building energy modeling
design
design for environment
energy
model validation
physical models
Physics-constraint learning
Physics-embedded algorithm structure
Physics-informed machine learning
prediction
Physics-embedded algorithm structure
Physics-informed machine learning
Building energy modeling
Physics-constraint learning
ISSN:0306-2619
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Building energy modeling (BEM) refers to computational modeling of building energy use and indoor dynamics. As a critical component in sustainable and resilient building development, BEM is fundamental to support a diverse spectrum of applications, including but not limited to sustainable building design and retrofitting, building resilience analysis, smart building control. Currently, two main approaches, i.e., physics-based and data-driven modeling, exist within BEM. Despite significant advancements of machine learning (ML) and deep learning (DL) algorithms in recent years, several challenges remain to apply these data-driven approaches in BEM, including the necessity of obtaining sufficient and high-quality training data in algorithm development, unreliable and physically infeasible predictions, and limited algorithm interpretability and generality in applications. These contribute to distrust and impede the widespread adoption of these algorithms in BEM practices. To overcome these challenges, this work provides a comprehensive overview of Physics-Informed Machine Learning (PIML), a novel modeling approach that encodes physics principles and useful physical information into cutting-edge ML algorithms. This approach is designed for advanced building energy modeling with enhanced robustness and interpretability. Specifically, existing PIML methods for BEM are summarized and categorized into different paradigms to integrate physics into ML models, including physics-informed inputs, physics-informed loss functions, physics-informed architectural design, and physics-informed ensemble models. The challenges, including the effective integration of prior physical knowledge in modeling and the evaluation of developed PIML methods, in the development of PIML for BEM are then discussed. This review outlines extensive existing research works and future potential research directions to shed light on the broader application of PIML to support BEM practice. •A comprehensive review of Physics-Informed Machine Learning (PIML) methods for Building Energy Modeling (BEM).•Present and discuss different methodologies of PIML application on BEM.•Discuss the challenges and potential solutions of PIML application in BEM.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:0306-2619
DOI:10.1016/j.apenergy.2024.125169