This study evaluates the performance prediction ability and model suitability of eleven empirically-based performance models for centrifugal water chillers. Specifically, this study uses over 2000 datasets with a constant or variable chilled water flow rate for fixed or variable speed drive centrifugal liquid chillers. The best regression coefficients for each empirical-based model were obtained using the ordinary least squares (OLSs) method. The model prediction accuracy of each empirical-based model is based on the coefficient of variation of root-mean-square error (CV). The evaluation for model suitability is based on the considerations of prediction ability, the complexity in training datasets, the effort needed to calibrate, the generality of the model, and its ability to physically interpret the model regression coefficients in this study. Results show that among the eleven empirical-based models, the BQ (CV=0.54%), MP (CV=0.61%), SMP (CV=0.70%), and MDOE-2 (CV=0.63%) models have overall prediction CV values under 1% for all kinds of datasets and achieve extremely good prediction accuracy. Because the MDOE-2 model has a more complicated datasets training process than the BQ, MP, and SMP models, and it has no ability to physically interpret the model regression coefficients, the BQ, MP, and SMP models have the best suitability. The results of this study provide important reference values for selecting empirically-based performance models for energy analysis, optimal operating control, energy efficiency measurement and verification (M&V), and the development of fault detection and diagnosis (FDD) systems in centrifugal water chillers.
