| Title |
Sampling Methods and Stochastic Inference in Monte Carlo Building Simulation |
| Authors |
Kim, Young-Jin ; Park, Cheol-Soo ; Kim, In-Han |
| Keywords |
Uncertainty Analysis ; Monte-Carlo ; Building Simulation ; Non-parametric Method ; Sampling Method |
| Abstract |
Building simulation tools have widely become available for prediction and assessment of building energy performance. However, in process of building simulation, numerous assumptions, simplifications of reality and uncertain inputs are required. Recently, a stochastic approach is being recognized as an alterative to deal with the aforementioned issues. In order to properly apply one of the stochastic methods, a Monte-Carlo approach to building simulation, this paper addresses the following: investigating applicability of the Quasi-Random Sampling and of the non-parametric Monte-Carlo method. In this paper, a range of fifty unknown parameters were selected and identified based on the literature, and then a screening method was applied to identify dominant inputs on simulation outputs. To verify the applicability of the Quasi-Random Sampling, this paper used the two sample Kolmogorov-Smirmov test that compares the differences between two samples (Quasi-Random vs. Simple Random, Quasi-Random vs. LHS [Latin Hypercube Sampling]). The paper shows that the Quasi-Random Sampling method is surprisingly good enough. The authors compared the non-parametric Monte-Carlo method (Kernel Density Function) with several parametric Monte-Carlo methods (Gamma, Exponential, Lognormal, Normal, Weibull, and Rayleigh distribution). The paper shows that the non-parametric Monte-Carlo method can reflect probabilistic characteristics of the reality in a building relatively well in comparison with the parametric methods. |