Explainable and Trustworthy Artificial Intelligence for Building Performance Diagnostics: Methods, Validation, and Reproducibility
Encyclopedia · 06.11.2026 · 6 (11) · 248
DOI: 10.3390/encyclopedia6110248 ↗Kurz gefasst
Fachartikel in der Zeitschrift Encyclopedia, online erschienen am 06.11.2026. Verfasst von José Iván San José Vieco. Die Arbeit nennt 61 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Instituto de Tecnologías Audiovisuales (ITAV), Universidad de Castilla-La Mancha (UCLM), Spain.
Abstract (Originalsprache)
Artificial intelligence (AI) is increasingly used for building energy prediction, anomaly detection, heating, ventilation, and air conditioning (HVAC) fault detection and diagnosis, retrofit assessment, and operational decision support. However, predictive performance alone does not establish whether an AI-based diagnostic system can be interpreted, transferred, reproduced, or used reliably in practice. This review examines methodological conditions supporting explainable and trustworthy AI for building performance diagnostics. A structured narrative review covered machine-learning and deep-learning applications, explainability methods, model validation, uncertainty quantification, cross-building generalization, reproducibility, climate-related distribution shift, and real-world deployment. The reviewed evidence shows that these dimensions address complementary aspects of diagnostic reliability. Explainability supports interpretation of model behavior but does not establish physical causality or diagnostic correctness; external validation is needed beyond development conditions; uncertainty becomes especially important under distribution shift; and transparent reporting supports reproducibility. Important gaps remain in longitudinal and cross-building validation, human-centered explanation evaluation, uncertainty-aware diagnostics, climate robustness, reproducible benchmarking, and long-term monitoring. An evidence-oriented workflow is proposed linking data provenance, model development, validation, uncertainty, explainability, documentation, human interpretation, operational action, and post-deployment monitoring, while distinguishing the evidence required for research-stage models, supervised decision support, and more autonomous operation.
Förderung
- Instituto de Tecnologías Audiovisuales (ITAV), Universidad de Castilla-La Mancha (UCLM), Spain
Zitieren
APA
Vieco, J. I. S. J. (2026). Explainable and Trustworthy Artificial Intelligence for Building Performance Diagnostics: Methods, Validation, and Reproducibility. Encyclopedia, 6(11), 248. https://doi.org/10.3390/encyclopedia6110248
Vancouver
Vieco JISJ. Explainable and Trustworthy Artificial Intelligence for Building Performance Diagnostics: Methods, Validation, and Reproducibility. Encyclopedia. 2026;6(11):248. doi:10.3390/encyclopedia6110248
BibTeX
@article{Vieco2026,
author = {Vieco, José Iván San José},
title = {Explainable and Trustworthy Artificial Intelligence for Building Performance Diagnostics: Methods, Validation, and Reproducibility},
journal = {Encyclopedia},
year = {2026},
volume = {6},
number = {11},
pages = {248},
doi = {10.3390/encyclopedia6110248}
}
EUVIMED