A comparative study of thermodynamic properties of R466A using linear regression, artificial neural network and gene expression programming

The use of next-generation refrigerant fluids is preferred to improve the global environment’s livability. In this context, the thermodynamic properties of R466A, a new-generation refrigerant with low ozone depletion potential and global warming potential, have been modelled using various methods. Linear regression, artificial neural network (ANN), and gene expression programming (GEP) models were used to predict R466A’s temperature–pressure relationship in the saturated liquid–vapor phase and its enthalpy-entropy relationship in the superheated vapor phase. The models’ performance was evaluated based on statistical parameters such as the determination coefficient (R²), mean absolute error, and root mean square error (RMSE), and compared with actual values. The research results indicate that the GEP model achieved the lowest RMSE values for predicting thermodynamic properties in the saturated vapor phase. On the other hand, ANN models were found to be more suitable for estimating properties in the superheated vapor phase. The R² values for ANN models ranged from 0.999 to 0.986, whereas GEP models exhibited R² values between 0.999 and 0.982. Despite slightly lower performance compared to some ANN models, GEP models employed explicit equations.