Machine Learning-Based Performance Modeling of Residential Heat Pump Systems Using Field Data

Heat pumps play a critical role in sustainable building operations, particularly in cold climates where efficient heating is essential. This study investigates the use of supervised machine learning models to predict the performance of a residential heat pump system, using real-world field data from the U.S. Department of Energy’s Heat Pump Field Data Repository.

The dataset, collected from a home in Massachusetts, includes outdoor air temperature, energy consumption, and heating demand recorded over time.

Despite the limited number of input features, models such as linear regression and decision trees demonstrated strong predictive capabilities. Feature engineering techniques were applied to enhance model accuracy and interpretability. Results show that outdoor temperature and user behavior are among the most influential factors affecting energy use and coefficient of performance.

This work demonstrates that predictive modeling is feasible even with sparse datasets and highlight the importance of thoughtful model selection and feature design. The findings support the application of data-driven methods in residential heating, ventilation, and air conditioning (HVAC) systems, where they can enhance control strategies, increase energy efficiency, and enable early detection of performance issues. This work contributes to national efforts to reduce carbon emissions and modernize residential energy systems through intelligent, adaptive technologies.