Integration of CFD and machine learning for application in water treatment process modeling: Membrane ozonation process evaluation

In this study, several tree-based machine learning models were developed and evaluated to predict the C (mol/m3) in membrane-based separation. The case study is membrane separation using ozonation for water treatment. Simulations were first conducted using computational fluid dynamics (CFD) to solve mass transfer equations and obtain concentration distribution of ozone in the process (C). Then the results were implemented in building machine learning models, thereby hybrid model was developed for correlation of solute concentration. The dataset consisted of 10,000 samples, each with two features of r (m) and z (m) which are the coordinates in radial and axial dimensions, respectively. Four models including Extra Trees (ET), Random Forest (RF), Gradient Boosting (GB), and Adaptive Boosted Trees (ADT) were trained and optimized using Firefly Algorithm (FA). The performance of each model was assessed using several metrics, including R-squared, mean squared error, mean absolute error, and maximum error. The results showed that all models performed well, with R-squared values ranging from 0.994 to 0.999 and maximum errors ranging from 0.144 to 0.639. Overall, the ADT model achieved the best performance, with an R-squared value of 0.999 and a maximum error of 0.143. These findings suggest that tree-based ensemble models can be utilized to accurately predict the C parameter in the separation process based on membrane.