Abstract
Uncertainty analysis quantifies the inherently uncertain nature of building energy performance, whereas sensitivity analysis identifies key factors to explain variations in building energy performance. With the ability to handle complex relationships, machine learning techniques offer an effective approach to more accurate and reliable uncertainty and sensitivity analysis. This paper provides valuable insights into the current state and future prospects of machine learning-based uncertainty and sensitivity analysis for building energy performance. The development of machine learning-based uncertainty analysis is discussed from three perspectives: observational data-based probabilistic prediction, surrogate model-based uncertainty quantification, and inverse uncertainty quantification. Variance-based sensitivity analysis using surrogate machine learning models decomposes output variance associated with each input. In contrast, machine learning-based variable importance refers to the change of model predictive performance using model-specific or model-agnostic approaches. Finally, future research directions on machine learning-based uncertainty and sensitivity analysis of building energy performance are presented.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data availability statement
No new data were created or analyzed during this study.