Artificial intelligence modeling for the determination of thermodynamic and transport properties of R454B refrigerant

Abstract Accurate knowledge of thermodynamic and transport properties is essential for optimizing the design and operation of refrigeration systems. With the continuous introduction of new refrigerants, there is an increasing demand for predictive approaches that can provide reliable property data over a broad range of temperatures and pressures, where experimental measurements are frequently unavailable. This study developed Deep Learning (DL), Gene Expression Programming (GEP), and Linear Regression (LR) models to predict the key properties of the low-GWP refrigerant R454B, such as enthalpy, entropy, specific volume, thermal conductivity, viscosity, and heat capacity. R454B has emerged as a promising, environmentally friendly alternative to R410A, so accurate prediction of its properties is crucial for the development of sustainable refrigeration technologies. Of the methods tested, the GEP model demonstrated the best performance, achieving R 2 values of over 0.99 for most properties and significantly reducing mean absolute error (MAE) and root mean square error (RMSE). These results highlight the robustness and sensitivity of GEP in capturing the complex thermophysical behavior of R454B. This study provides a novel and effective modeling framework that paves the way for more efficient, accurate, and sustainable prediction of refrigerant properties.