Stochastic Global Sensitivity Analysis for Retrofit Analysis in an Existing Building

Sensitivity analysis has been widely used for rational decision making of energy retrofit alternatives. A Sobol method is a variance based approach and provides powerful post-processing sensitivity results that enable to quantitatively identify influential and non-influential inputs. It can account for impacts of each input as well as impacts caused by all possible interactions between each input on building energy consumption. However, the Sobol method requires vast computation time. To overcome such computational disadvantage of the Sobol method, the authors employed a surrogate model using a Gaussian Process (GP) emulator. In other words, we combined the GP emulator and the Sobol method for energy retrofit decision making. In the paper, the following are addressed in detail: (1) global sensitivity analysis using the GP emulator, (2) decision making based on stochastic prediction.