Application of a probabilistic LHS-PAWN approach to assess building cooling energy demand uncertainties

A deterministic approach to building energy simulation risks the omission of real-world uncertainties leading to prediction errors. This paper highlights limitations of this approach by contrasting it with a probabilistic uncertainty/sensitivity simulation approach. Latin hypercube sampling (LHS) generates 15000 unique model configurations to assess the effects of weather, physical and operational uncertainties on the annual and peak cooling energy demands for a residential building which situated in a hot and dry climatic region. Probabilistic simulations predicted 0.22-2.17 and 0.45-1.62 times variation in annual and peak cooling energy demands, respectively, compared to deterministic simulation. A novel density-based global sensitivity analysis (SA), i.e., PAWN, is adopted to identify dominant input uncertainties. Unlike traditional SA methods, PAWN allows simultaneous treatment of continuous and categorical inputs from a generic input-output sample. PAWN is favourable when computational resources are limited and model outputs are skewed or multi-modal. For annual and peak cooling demands, the effects of weather and operational parameters associated with airconditioner and window operation are much stronger than these of other parameters considered. Consequently, these parameters warrant greater attention during modelling and simulation stages. Bootstrapping and convergence analysis also confirm the validity of these results.

Automated summary

This study contrasts deterministic approach in building energy simulation with probabilistic uncertainty/sensitivity simulation approach, highlighting the higher accuracy and reliability of the latter. The PAWN method allows simultaneous treatment of continuous and categorical inputs, and is favorable when computational resources are limited and model outputs are skewed or multi-modal.