A Data-Driven Framework for Estimating Residential Energy Savings: Machine Learning Applications for Low-Income Housing
Date of Award
12-12-2024
Degree Name
M.S. in Renewable and Clean Energy Engineering
Department
Department of Mechanical and Aerospace Engineering
Advisor/Chair
Kevin Hallinan
Abstract
This study presents a data-driven model designed to estimate energy savings for residential buildings, with a focus on low-income communities where energy efficiency improvements hold significant potential. The model integrates machine learning, nearest neighbor analysis, and clustering techniques to address limitations in traditional energy savings methods, which often lack the adaptability and regional specificity required for accurate predictions. Key objectives include the development of predictive models using building characteristics, refinement of savings estimates through Coefficient of Variation (CoV) analysis, and the creation of a graphical user interface (GUI) that provides accessible energy savings estimates for end-users, such as city planners and residents. The model was trained on the National Renewable Energy Laboratory (NREL) dataset and further validated through clustering and statistical analysis to ensure reliability across diverse energy-saving scenarios, including enhancements in insulation, infiltration, HVAC efficiency, and thermostat settings. Results demonstrate that clustering enhances prediction stability by filtering out high-variability data points, while CoV analysis aids in prioritizing buildings with the highest potential savings, especially in high-energy-use or low-income contexts. Findings indicate that the model effectively identifies high-impact energy efficiency measures, guiding resources toward modifications that promise the most substantial reductions in energy consumption. Although the results provide general estimates rather than precise savings values, they offer actionable insights into which building improvements are likely to be most effective, accompanied by metrics that communicate prediction reliability. The accessible GUI developed as part of this study enables end-users to retrieve address-specific savings predictions, promoting data-informed decision-making for energy efficiency improvements. This research contributes a practical tool for residential energy planning, supporting stakeholders in advancing sustainability and reducing energy costs in vulnerable communities.
Keywords
machine learning, energy savings, energy efficiency improvements, nearest neighbor analysis, clustering, low-income houses
Rights Statement
Copyright © 2024, author.
Recommended Citation
Clayton, Philip, "A Data-Driven Framework for Estimating Residential Energy Savings: Machine Learning Applications for Low-Income Housing" (2024). Graduate Theses and Dissertations. 7467.
https://ecommons.udayton.edu/graduate_theses/7467
