Developing a semi-empirical model for refrigerant compressors using symbolic regression: a data-driven approach for heat pump applications
Accurate thermodynamic modeling of compressors is essential for the analysis, optimization and control of heat pump systems. Traditional modeling methods, such as the AHRI-540 10-coefficient polynomial, lack generalizability and are fluid-specific, requiring costly re-parameterization for different refrigerants. To address this limitation, this work explores symbolic regression (SR) as a transparent, data-driven technique to derive a refrigerant-generalized correlation for isentropic efficiency. This methodology was applied to a comprehensive experimental dataset from reciprocating compressors operating with a diverse range of fluids, including HFOs, HFCs, HCs and their mixtures. The SR search successfully discovered a novel, parsimonious correlation for isentropic efficiency that is a function of these physical properties. However, the current correlation still has limitations when it comes to correctly predicting performance over the entire compressor envelope, particularly in areas where there is little measuring data available. This paper presents the methodology, the proposed correlation and a discussion of its benefits and limitations. The physics-informed SR approach offers a promising path to generalizable models, though the proposed equation requires further refinement to achieve full topological accuracy across more refrigerant classes.