Using Machine Learning (ML) for Heat Transfer Coefficient (HTC) measurement in buildings: A systematic review

Accurate, fast, and non-intrusive Heat Transfer Coefficient (HTC) estimations of existing buildings are crucial for informed decision-making in retrofit projects. A systematic integrative literature review was conducted following the PRISMA standard (1997–2024). This is the first review with statistical and critical thematic analysis on this topic, synthesizing a theoretical framework for future research directions. The review identified 63 relevant sources.

Quantitative analysis reveals exponential growth of publications since 2017 on ML for HTC estimation, with leadership from China, USA and India. Qualitative analysis assessed existing ML methodologies, data collection protocols, and validation frameworks. Indoor and outdoor temperatures emerge as the most frequently used parameters; 45 % of studies use hourly measurement intervals. High-frequency data collection offers superior capture of thermal dynamics but poses challenges in data storage and processing. Neural network-based approaches (particularly multilayer perceptrons (MLP) and long short-term memory (LSTM)) dominate to capture non-linear thermal relationships. Key challenges include data quality requirements, computational efficiency, and the trade-off between model complexity and interpretability. A concerning finding shows 25.7 % of studies lack validation against established methods, highlighting a critical gap in standardization.

This analysis provides insights into current methodological trends, identifies research gaps, and proposes future directions for advancing ML applications in this field. Findings emphasize the need for standardized validation protocols, improved data collection strategies, and integrated approaches connecting component-level and whole-building analyses.