Evaluating the efficiency of nanofiltration and reverse osmosis membranes for the removal of micro-pollutants using a machine learning approach

Abstract

Water removal research is a critical study area across diverse industries and scientific domains. This investigation evaluates three prominent machine learning models including Artificial Neural Networks, Random Forest, and XGBoost for their predictive efficacy concerning water removal efficiency in reverse osmosis (RO) and nanofiltration processes. The study aims to assess the influence of molecular weight (MW) and compound size on removal rates within RO and nanofiltration contexts. After training with an extensive dataset, the Artificial Neural Network (ANN) model was created using a four-layer architecture with 128, 64, 32, and 32 nodes over 2500 epochs. Notably, the ANN model exhibited superior performance, achieving an R2 score of 0.9795, indicating its precision in elucidating the variability observed within RO and nanofiltration processes. The Random Forest model (R2: 0.9658) and XGBoost model (R2: 0.9447) also demonstrated substantial predictive capabilities. These results offer valuable insights for optimizing water removal techniques within these processes.

Keywords

• Machine learning models
• Artificial neural networks (ANN)
• Reverse osmosis (RO)
• Nanofiltration
• Efficiency