Paper Conference

Proceedings of ASim Conference 2012: 1st Asia conference of IBPSA-China, Japan, Korea


Bayesian calibration for transient energy simulation model

Dong-Hyun Lee, Young-jin Kim, Cheol-soo Park

Abstract: Building simulation has become increasingly important because of its capability to assess potential energy efficiency savings in buildings. In the simulation modellingprocess, many assumptions and simplifications are required and many uncertain inputs are also involved. The aforementioned issues may cause a significant discrepancy between a reality and prediction. With this in mind, this paper investigatesapplicability of a Bayesian calibration technique to a transient simulation model. This will improve prediction accuracy and reduce uncertainty. The Bayesian calibration is a way to estimate the posterior distribution based on the quantified prior distribution.For sampling uncertain inputs, MCMC(Markov Chain Monte Carlo) method was applied in our study. One of the MCMC methods, DRAM(Delayed Rejection Adaptive Metropolis) was employed. This paper addresses Bayesian calibration, uncertainty analysis, and risk analysis for whole-building building energy simulation modelling(EnergyPlus). For this study, an office building was selected and 28 unknown inputs were identified. The Bayesian calibration was conducted in three steps: (1) determination of prior probability distributions for uncertain inputs, (2) formulation of likelihood functions,and (3)MCMC method for posterior distributions. Lastly, Coefficient of Variation of the Root Mean Squared Error (CVRMSE, ASHRAE Guide line14) was used for the validation of the approach. In the paper, the following is discussed: (1) posterior distributions of inputs against their prior distributions, (2)results of Bayesiancalibration for the model, and (3) advantages of Bayesian calibration for dynamic simulation model.
Pages: 426 - 433