An evaluation of empirically-based models for predicting energy performance of vapor-compression water chillers

This paper presents an evaluation of six empirically-based models for predicting water chiller energy performance using over 1000 chiller data sets from chiller manufacturers and field measurements. The data sets comprise three broad classifications, including (1) constant condenser and constant chilled water flow, (2) constant condenser and variable chilled water flow, and (3) variable condenser and variable chilled water flow. The regression parameters for each performance model are obtained using least squares method. The criteria for evaluating the predictive ability of models are based on the coefficient of variation of root-mean-square error (CV). Results show that among the six empirically-based performance models for water chillers in this study, the bi-quadratic regression model (CV = 2.2%) and the multivariate polynomial regression model (CV = 2.25%) have the best prediction accuracy for all kinds of data sets. The results of this study can be used as a reference for selecting empirically-based models for the purposes of energy analysis, performance prediction, evaluation of energy-efficiency improvements, and fault detection and diagnosis of water chillers.