Conference Paper

Proceedings of SimBuild Conference 2024

Uncertainty Propagation in Building Analysis with Truncated Taylor Polynomials

Richard Walter Fenrich
xStar Research, United States of America

Abstract: We propose a non-probabilistic method for propagating uncertainties represented as intervals through a building performance simulation. The method only requires a single simulation without any prior computation. It is based on Taylor model arithmetic, and solves previous issues with interval-based methods such as the dependency problem by using truncated Taylor polynomials. As a result, the method provides an approximate estimate of uncertainty. However, initial results compare favorably in accuracy to Monte Carlo. Results are shown for a cubic function with 10,000 uncertain variables, and transient thermal simulations of a box with static and dynamic external boundary conditions. The proposed method is promising for efficient non-probabilistic uncertainty quantification in building performance simulation and warrants further investigation and refinement.
Keywords: uncertainty quantification, interval arithmetic, Taylor model arithmetic, energy model
Pages: 148 - 158
Paper: