Artificial intelligence modeling and simulation of membrane-based separation of water pollutants via ozone Process: Evaluation of separation

Abstract

The present study offers a comparative examination of regression models that are utilized for the prediction of concentration (C) in a new hybrid ozone-membrane process for removal of water pollutants. The main focus is on the tracking ozone concentration in the feed side of a membrane contactor system. Computational fluid dynamics (CFD) was carried out to obtain data of ozone concentration (C) for developing some machine learning (ML) models. The models are based on input variables r (m) and z (m). The dataset comprises over 10,000 data, and three different models, namely Convolutional Neural Network (CNN), Support Vector Regression (SVR), and Orthogonal Matching Pursuit (OMP), are evaluated. The hyperparameters of these models are optimized using the Glowworm Swarm Optimization (GSO) technique. Prior to model training, preprocessing steps are applied. The findings suggest that SVR exhibited a noteworthy R2 score of 0.99698, surpassing CNN which obtained a R2 score of 0.98073, and OMP which obtained a R2 score of 0.8748. The aforementioned discoveries offer significant perspectives on the utilization of diverse machine learning models in the prognostication of C, demonstrating their efficacy and proficiency in this particular realm.