The optimization of sensitivity coefficients for the virtual in situ sensor calibration in a LiBr–H₂O absorption refrigeration system

The correct data or information from the building sensing networks plays a vital role in the operation algorithms. The sensor errors usually show a negative effect on the performance of control, diagnosis, and optimization of building energy systems. Thus, the physical working sensors periodically need to be removed to be calibrated by the reference sensors, which will disrupt the normal operation of building systems from time to time. The virtual in situ sensor calibration (VIC), based on the Bayesian inference and Markov chain Monte Carlo methods (MCMC), is an effective approach to handle the systematic and random errors of various working sensors simultaneously. This technology uses the distance function and system models to estimate the true measurements and addresses most of the practical problems in a traditional calibration process. However, the sensitivity coefficient in the definition of distance function is one of the determining factors in the calibration accuracy and how to define it still remains uncertain. Therefore, this study employed the genetic algorithm (GA) to optimize this parameter in a LiBr–H₂O absorption refrigeration system. The results revealed that the systematic and random errors of temperature and mass flow rate were reduced considerably with the help of optimized sensitivity coefficients and most of the measurements approached to their true values after the calibration.