Solubility Prediction of Refrigerants in PEC Lubricants Based on Back-Propagation Neural Network Combined with Genetic Algorithm

In the present study, a backpropagation neural network combined with genetic algorithm (GA-BP) model was constructed for prediction the solubility of refrigerants in linear chained precursors of POE lubricants (PECs). A total of 2248 experimental solubility data of refrigerants in PECs reported in literature were collected with temperatures from 243.15 K to 363.15 K and pressures up to 10 MPa. The input variables of the model were optimized using non-dominated sorting genetic algorithm with elite strategy (NSGA-II). The optimized inputs include temperature, pressure, molecular weight, critical temperature, and acentric factor. Results indicate that the GA-BP model using the optimized inputs can correlate the solubility data with good accuracy, the average absolute relative deviation between calculated results from the model and the literature is 0.98 %. Moreover, in order to validate the predictive ability of the established GA-BP model, the solubility of R1243zf in PEC4 and PEC5 was measured at the temperature range from 278.15 K to 343.15 K. The calculated values from the GA-BP model were compared with the experimental data, and the average absolute relative deviation is 8.85 %. Finally, sensitivity analysis was performed by Partial Derivatives (PaD) method to assess the contribution of input variables. Leverage approach was used for outlier detection to ensure the robustness of the model.