Molecular separation and computational simulation of contaminant removal from wastewater using zirconium UiO-66-(CO₂H)₂ metal–organic framework

We proposed a methodology based on machine learning approach for estimation of ion separation via adsorption technique. The case study considered in this work is removal of two water pollutants including Hg or Ni from water using a metal organic framework (MOF) material, with the formula of UiO-66-(Zr)-(COOH)₂. A set of data was collected from literature and then used for training and validation of the model. The used data have two outputs for Qe and Ce which are the adsorption capacity and the equilibrium concentration, respectively. The modeling takes two inputs: ion type (Hg or Ni) and initial ion concentration (C₀). We analyzed and modeled the data employing three different regression models, including multilayer perceptron (MLP), linear support vector regression (LSVR), and Gaussian process regression (GPR), to make regression on this data. Implementation and testing of the final models followed the tuning of hyper-parameters using SMA algorithm. With R² criterion, three models were shown the score of more than 0.92 for both Ce and Qe. Despite the fact that all models have acceptable performances, GPR has been shown to have the largest generality and accuracy for both outputs. As a result, it was selected as the main model in our study. For Ce and Qe, the RMSE metrics calculated using GPR are 4.22E + 00 and 4.98E-01, respectively, based on the GPR.