Data-driven optimisation of residential air-to-water heat pump performance using IoT and machine learning

Residential heating accounts for about 27 % of the UK’s energy consumption. While residential heat pumps (RHPs) are central to the transition toward sustainable energy, optimising their real-world performance requires robust experimental monitoring and predictive modelling. This study presents a data-driven approach for evaluating and optimising the performance of residential air-to-water heat pumps (A2WHPs) using real-time data and machine learning (ML). A full-scale experimental setup was deployed in a UK-based end-terrace building, incorporating IoT-enabled sensors to capture 275 days of operational data that was processed into a 6,600-hour dataset. Key thermal, electrical, and environmental parameters were measured at high temporal resolution and used to develop predictive models for the system’s coefficient of performance (COP). Several ML models, including Random Forest, Support Vector Regression (SVR), eXtreme Gradient Boosting (XGBoost), Artificial Neural Networks (ANN), and Long Short-Term Memory (LSTM), were evaluated using rigorous preprocessing, principal component analysis, and GridSearchCV hyperparameter tuning. LSTM, XGBoost, and ANN achieved the highest prediction accuracy with low error values across MAE, MSE, RMSE, CVRMSE, and NMBE. Diagnostic plots and residual analysis further confirmed the generalisability of the models and their sensitivity to non-linear operational behaviours. The findings demonstrate that integrating ML with real-world data can provide a robust predictive framework for operational diagnostics, performance evaluation, and efficiency improvement in residential heat pumps. This approach supports scalable, data-driven energy management and contributes to decarbonising the built environment.