To meet the carbon emissions reduction targets set under the Paris Agreement and effectively combat climate change, transitioning to sustainable heating practices in residential buildings is essential. Heat pumps are a pivotal technology in this effort to provide domestic hot water and space heating with high energy efficiency. The International Energy Agency projects that up to 600 million heat pumps will need to be installed globally by 2030 to achieve a net-zero scenario, potentially reducing greenhouse gases equivalent to the current annual emissions of all cars in Europe.
However, several challenges hinder the widespread adoption and effective operation of heat pumps, causing many countries to fall short of their installation targets. Currently, global installation rates indicate a 58% shortfall of the targets needed to achieve the net-zero scenario. One major challenge is the significant shortage of trained professionals capable of installing and maintaining heat pumps. Additionally, the replacement of fossil fuel-based heating systems with heat pumps involves substantial costs. Many homeowners are reluctant to invest in this technology due to insufficient and frequently changing policy frameworks. Another challenge is the variability in real-world heat pump performance, which often diverges from the efficiency levels reported on product certificates, resulting in unmet expectations regarding operational costs. Furthermore, many users and installers lack familiarity with heat pump technology, which differs significantly from traditional gas and oil boilers that have been optimized over decades. This knowledge gap can lead to systems that are improperly sized, malfunctioning, or misconfigured, and there is a lack of effective feedback mechanisms for performance assessment. Finally, the widespread installation of heat pumps can place added strain on power grids, increasing both peak and total electricity demand, especially if the systems are not operating efficiently.
This thesis addresses the critical issues in heat pump adoption and operation by investigating how energy data from widely deployed smart electricity meters and sensor data from modern, internet-connected heat pumps can be leveraged to optimize performance. Through a series of four articles, it develops practical, machine learning-based methodologies for evaluating the performance of residential heat pumps post-installation and identifying systems with energy-saving potential. Focusing on heat pumps used for heating applications in Central Europe, the thesis further provides data-driven insights into real-world operation. The methodologies and findings can directly be applied to energy efficiency services, delivering personalized feedback on heat pump performance to guide installers, users, and other stakeholders.
The first article develops methods for classifying heat pumps into efficiency categories, assessing system sizing, and estimating energy-saving potential through configuration adjustments using sensor data. Analysis of field data from 1,023 heat pumps over two years shows that 11% are incorrectly sized, with 17% of air-source and 2% of ground-source heat pumps failing to meet efficiency standards. The second article introduces a method for distinguishing heat pumps based on their modulation capabilities using 15-minute resolution smart meter data. A K-nearest neighbor algorithm effectively differentiates between fixed speed and variable speed systems, achieving an AUC of 0.976 with just one week of data, where the heat pump is measured alongside other appliances. The third article introduces a deep learning algorithm designed to isolate heat pump patterns from aggregated 15-minute energy measurements recorded by a smart meter. Tested with data from 363 households, the algorithm achieves an average RMSE of 0.169 kWh, comparable to the accuracy of previous studies utilizing high-resolution measurements. The fourth article develops an algorithm to extract key indicators of heat pump on-off behavior from smart meter data and identify atypical systems. Analysis of 503 Swiss households over 21 months shows that cycling behavior is relatively consistent across different building and heat pump characteristics. However, outliers in cycling behavior are more than twice as common in air-source heat pumps compared to ground-source heat pumps.
In addition to its technical contributions and data-driven insights, this thesis offers a comprehensive overview of emerging trends and service opportunities at the intersection of heat pump technology and digitalization. It examines the shift from a fragmented heat pump market to a more consolidated structure dominated by a few major players, and discusses how this consolidation might impact prices, demand, and the potential entry of new competitors. The thesis also explores the transition in digital services from standalone monitoring to integrated optimization solutions that consider all building components. While energy efficiency monitoring will remain crucial, the focus is expected to shift towards leveraging heat pump operational flexibility to capitalize on dynamic electricity prices and lower operational costs. This will likely involve advanced machine learning algorithms for optimization and adaptive control, potentially running locally on newer heat pump models rather than in the cloud. Finally, the thesis anticipates that new standards may mandate heat pump monitoring and control, driving technical standardization and facilitating the scaling of services across various manufacturers and models.
