A quantitative comparison of statistical and deterministic methods on virtual in-situ calibration in building systems

Systematic and random errors of working sensors in building systems could significantly compromise the system's performance and thus indoor environmental quality. An extended virtual in-situ calibration has been suggested to solve problems regarding sensor errors and calibration. This calibration can correct these errors for all critical working sensors in building systems without removing working sensors or adding reference sensors as is done in a conventional calibration. This method is capable of estimating measurands using a parameter estimation technique based on mathematical system models. Deterministic and statistical methods can be used for conducting the estimation. In this study, genetic algorithm (GA)-based optimization is used as a deterministic method and Bayesian Markov Chain Monte Carlo (MCMC) is used as a statistical method to solve the calibration problem formulated by the extended virtual in-situ calibration. A case study of a single-effect LiBr-H₂O refrigeration system illustrates the problem formulating process and compares the accuracy distributions of calibrations derived from the two different methods.