Urbanization trends have intensified the focus on predicting building energy consumption within urban areas for sustainable development. Urban Building Energy Modeling (UBEM) offers a valuable approach to simulating and evaluating building energy efficiency within urban contexts, considering various physical and climatic factors. This paper explores the application of data-driven UBEM in urban energy planning, with the case study of Turin, Italy. It is due to the fact that traditional physics-based UBEM models face limitations in large-scale urban settings, prompting the adoption of data-driven approaches. The study evaluates the effectiveness of Machine Learning (ML) algorithms, particularly Light Gradient-Boosting Machine (LightGBM) and Random Forest (RF), in predicting energy consumption for space heating at both monthly and hourly time steps. Using a comprehensive dataset of 44,290 buildings and building blocks and the District Heating Network (DHN) in Turin with 6146 connected buildings, the study demonstrates the superior predictive performance of LightGBM over Random Forest, particularly at the urban scale. In the stable operational months from December 2022 to March 2023, LightGBM showed a maximum relative error of 2% for monthly energy consumption prediction, while RF had a maximum relative error of 9%. For buildings’ hourly energy consumption profile, despite challenges associated with space heating cut-off during a day, both algorithms exhibit robust performance, with relative errors below ±20% for most of the hours. These results highlight the robustness of both ML models in accurately predicting monthly energy consumption, particularly for urban application.

Machine Learning algorithms for Urban Building Energy Modeling / Montazeri, Ahad; Mutani, Guglielmina. - ELETTRONICO. - 12th International Conference on Improving Energy Efficiency in Commercial Buildings and Smart Communities:(2024), pp. 48-62. (Intervento presentato al convegno 12th International Conference on Improving Energy Efficiency in Commercial Buildings and Smart Communities tenutosi a Frankfurt, Germany nel 6-7/03/2024) [10.2760/716916].

Machine Learning algorithms for Urban Building Energy Modeling

Ahad Montazeri;Guglielmina Mutani
2024

Abstract

Urbanization trends have intensified the focus on predicting building energy consumption within urban areas for sustainable development. Urban Building Energy Modeling (UBEM) offers a valuable approach to simulating and evaluating building energy efficiency within urban contexts, considering various physical and climatic factors. This paper explores the application of data-driven UBEM in urban energy planning, with the case study of Turin, Italy. It is due to the fact that traditional physics-based UBEM models face limitations in large-scale urban settings, prompting the adoption of data-driven approaches. The study evaluates the effectiveness of Machine Learning (ML) algorithms, particularly Light Gradient-Boosting Machine (LightGBM) and Random Forest (RF), in predicting energy consumption for space heating at both monthly and hourly time steps. Using a comprehensive dataset of 44,290 buildings and building blocks and the District Heating Network (DHN) in Turin with 6146 connected buildings, the study demonstrates the superior predictive performance of LightGBM over Random Forest, particularly at the urban scale. In the stable operational months from December 2022 to March 2023, LightGBM showed a maximum relative error of 2% for monthly energy consumption prediction, while RF had a maximum relative error of 9%. For buildings’ hourly energy consumption profile, despite challenges associated with space heating cut-off during a day, both algorithms exhibit robust performance, with relative errors below ±20% for most of the hours. These results highlight the robustness of both ML models in accurately predicting monthly energy consumption, particularly for urban application.
2024
978-92-68-14947-8
File in questo prodotto:
File Dimensione Formato  
JRC137722_01.pdf

accesso aperto

Tipologia: 2a Post-print versione editoriale / Version of Record
Licenza: Creative commons
Dimensione 5.61 MB
Formato Adobe PDF
5.61 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/2989359