Optimizing uranium removal from groundwater using machine learning models for enhancing the uranium-adsorption process

초록

Uranium, a naturally occurring element in groundwater, poses significant risks to the environment and public health. Recent studies have reported efficient uranium adsorption from groundwater by heated aluminum oxide particles (HAOPs). However, optimizing the adsorption efficiency is challenging because of the complexity of the process due to various factors, including the adsorbent dosage, pH, uranium concentration, and flow rate relative to the specific volume. Repeating experiments to achieve optimization is time-consuming, expensive, and labor-intensive. Thus, this study applied machine learning (ML) modeling techniques including multilayer perceptron (MLP), support vector regression (SVR), extreme gradient boosting (XGB), decision tree (DT), and random forest (RF) models to optimize uranium adsorption by HAOPs. Important features were analyzed using Shapley additive explanations (SHAP) to quantify the relative importance of each feature in the best model. Experimental results show increased adsorption at high adsorbent dosages, although it decreased with increasing pH and uranium concentration. However, flow rate did not exhibit a significant effect. XGB performed better than RF and MLP, with an R-2 of 0.951, and outperformed DT and SVR in predicting uranium adsorption by HAOPs. SHAP results reveal the following order of feature importance for the XGB model: specific volume > uranium concentration > adsorbent dosage > pH > flow rate. Overall, this study demonstrates the effective use of ML techniques, specifically XGB, for predicting uranium adsorption by HAOPs, offering benefits in terms of reduced experimental costs, workload, and time consumption.

키워드