Staf Roels

Assessing the overall Heat Transfer Coefficient (HTC) of a building received growing attention to gain insights into the energy performance gap of our building stock. So far, though, there is a lack of clear guidelines on optimal conditions for reliable in-use HTC assessment. The current study explores how in-use HTC-measurements vary with weather conditions, monitoring duration, and data granularity, focusing on two widely used methods: Multiple Linear Regression (MLR) and Auto-Regressive with Exogenous Input (ARX) modelling. Both techniques are applied to a comprehensive dataset of dwellings equipped with smart meters for gas and electricity, along with indoor temperature data from nearly all rooms, collected over one to three years. The extended monitoring period enabled analysis of HTC evolution and its standard deviation over time, helping identify optimal weather conditions, monitoring equipment and monitoring duration. The most reliable values were achieved in winter conditions under high indoor-outdoor temperature differences and low solar radiation. ARX methods demonstrated greater robustness throughout the year compared to MLR-methods. Reducing the number of indoor temperature sensors was found to significantly affect the estimates.

To address the energy performance gap in buildings, research has increasingly focused on assessing actual energy performance, particularly through the overall Heat Transfer Coefficient (HTC), a metric reflecting a building’s as-built fabric efficiency. Traditional HTC determination relied on controlled yet intrusive measurements. Recently, the focus shifted toward in-use and on-board methods, which are minimally or non-invasive but offer less control. The advantage of these newer approaches is their scalability, enabling assessments to be conducted across a larger number of buildings.,This paper evaluates the applicability of in-use methods by comparing the common statistical approaches currently in place to determine a building’s HTC. These methods, ranging from simple static linear regression techniques to more advanced dynamic models, are applied to gas and electricity metering in combination with indoor temperature data collected from a large sample of houses. While most methods yield comparable results, those excluding solar radiation tend to underestimate HTC. Methods incorporating solar gains produce higher values but vary in reliability. ARX models show year-round robustness, whereas multiple linear regression (MLR) models perform best under winter conditions with extended monitoring. Sensor reduction significantly affects accuracy, with deviations up to 25%. The study also explores ARX weighting strategies and challenges the assumption of a unique HTC-value by examining wind effects, offering a first step toward diagnosing underlying heat loss mechanisms. These findings highlight the strengths and limitations of current methods and suggest directions for future research in building performance diagnostics.