New method based on neuro-fuzzy system and PSO algorithm for estimating phase equilibria properties

• Published in: Chemical Industry & Chemical Engineering Quarterly Vol. 28; no. 2; pp. 141 - 150
• Main Authors: Hadj, Abdallah, Laidi, Maamar, Hanini, Salah
• Format: Journal Article
• Language: English
• Published: Belgrade Association of the Chemical Engineers of Serbia 01.01.2022
• Subjects: Acids; Algorithms; anfis; Artificial intelligence; Artificial neural networks; Back propagation networks; Carbon dioxide; critical properties; Drugs; Equations of state; Equilibrium; Fuzzy logic; Fuzzy sets; Learning; modeling; Neural networks; Optimization algorithms; Optimization techniques; particle swarm optimization; Pharmaceuticals; Phase equilibria; Regression analysis; Solubility; Statistical analysis
• ISSN: 1451-9372; 2217-7434; 2217-7434
• DOI: 10.2298/CICEQ201104024A

The subject of this work is to propose a new method based on the ANFIS system and PSO algorithm to conceive a model for estimating the solubility of solid drugs in supercritical CO2 (sc-CO2). The high nonlinear process was modeled by the neuro-fuzzy approach (NFS). The PSO algorithm was used for two purposes: replacing the standard backpropagation in training the NFS and optimizing the process. The validation strategy has been carried out using a linear regression analysis of the predicted versus experimental outputs. The ANFIS approach is compared to the ANN in terms of accuracy. Statistical analysis of the predictability of the optimized model trained with a PSO algorithm (ANFIS-PSO) shows a better agreement with the reference data than the ANN method. Furthermore, the comparison in terms of the AARD deviation (%) between the predicted results, the results predicted by the density-based models, and a set of equations of state demonstrates that the ANFIS-PSO model correlates far better with the solubility of the solid drugs in scCO2. A control strategy was also developed for the first time in the field of phase equilibrium by using the neuro-fuzzy inverse approach (ANFISi) to estimate pure component properties from the solubility data without passing through the GCM methods.