Exploring structure–property relationships of critical temperatures for binary refrigerant mixtures via group contribution and machine learning

Thermodynamic cycles are the main approach of energy conversion, which is the main source of carbon emission. The working fluid is the energy carrier of thermodynamic cycles. And refrigerant is widely employed in low and medium grade energy utilization and heating ventilation and air conditioning. The refrigerant mixtures can effectively combine the advantages of their components, which plays a key role in decarbonization. As a basic thermophysical property, critical temperature, T_c, plays an important role in thermodynamic calculation and thermodynamics system design. In this work, the structure-property relationship models of T_c for binary refrigerants were established by developing predictive models based on 61 binary refrigerants with 275 sets of experimental T_c data and six machine learning algorithms. Also, specific halogenated groups of refrigerants are used to characterize the components and molecular structures of binary mixtures. The Multiple-layer Perceptron model owns the best fitting and generalization ability with the average deviation is lower than 2 ​%. Compared with conventional methods, the proposed model does not rely on any experimental property data or empirical parameters, and can accurately predict T_c of binary refrigerant mixtures directly from their components and mixing ratios. The present work could be guided in building predictive models for other properties, thereby supporting the development of novel refrigerant mixtures.