Towards a hybrid retrofit planning framework: a Data-driven tool for energy retrofit in residential buildings

Seraj, Hamidreza, Abbaspour, Atefeh and Bahadori-Jahromi, Ali ORCID logoORCID: https://orcid.org/0000-0003-0405-7146 (2025) Towards a hybrid retrofit planning framework: a Data-driven tool for energy retrofit in residential buildings. Energy and Built Environment.

[thumbnail of Towards a hybrid retrofit planning framework- May 2025.pdf] PDF
Towards a hybrid retrofit planning framework- May 2025.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Attribution Non-commercial.

Download (2MB)

Abstract

Since only less than one percent of the existing housing stock is replaced annually, retrofitting existing buildings has become the most effective approach to meet energy efficiency targets in this sector. Recently, researchers have promoted AI data-driven tools over traditional physics-based parametric simulations for retrofit planning; however, most studies rely on statistical metrics to evaluate performance of developed models, lacking comparative analysis with parametric simulation results across diverse case studies to evaluate practical applicability and identify potential limitations. This study aims to achieve three key objectives: first, to develop a novel web-based data-driven tool with advanced capabilities, including predicting residential buildings' energy consumption, CO₂ emissions, and EPC rating; evaluating the impact of various retrofitting measures, considering renewable energy solutions, on energy performance and environmental outcomes; and estimating retrofit costs based on real-time market data. Second, it conducts a comparative analysis with parametric simulation results for a case study in the UK to demonstrate the potential of a hybrid framework in supporting energy retrofit planning. Lastly, the study identifies optimal retrofit solutions by integrating energy, economic, and environmental considerations. The research utilises the XGBoost machine learning algorithm for developing the data driven tool, along with techniques such as SMOTE to address data imbalance and GA optimisation for hyperparameter tuning. Furthermore, EDSL-TAS is used for the parametric simulation of the case study buildings. The results demonstrate that the developed data-driven tool achieves over 81% accuracy in predicting CO2 emissions and energy consumption, and approximately 79% accuracy for the EPC rating. Further comparative analysis with parametric simulation results for a typical UK house revealed less than 5% error in predicting energy consumption for building envelope retrofits. Among the analysed retrofit scenarios, replacing the existing system with an ASHP appeared as the optimal final choice due to its best performance in reducing the energy consumption (57% reduction) and CO₂ emissions (77% reduction). The developed data-driven tool offers a reliable, time-efficient alternative to simulation models, providing accurate cost and impact assessments. It is user-friendly for non-professionals and home-owners without requiring simulation expertise and allows professionals to evaluate retrofit costs and prioritise options for detailed simulation analysis and therefore, enhancing decision-making.
Keywords:
Machine learning; Data-driven model; Energy efficiency; Building retrofit; Carbon emissions

Item Type: Article
Identifier: 10.1016/j.enbenv.2025.05.013
Keywords: Machine learning; Data-driven model; Energy efficiency; Building retrofitCarbon emissions
Subjects: Computing
Depositing User: ALI BAHADORI-JAHROMI
Date Deposited: 28 May 2025 14:30
Last Modified: 28 May 2025 14:45
URI: https://repository.uwl.ac.uk/id/eprint/13700
Sustainable Development Goals: Goal 7: Affordable and Clean Energy Sustainable Development Goals: Goal 11: Sustainable Cities and Communities

Actions (login required)

View Item View Item

Menu