Intelligent Models as Novel Tools for Optimizing Ultrasonication-Ozonation Technique in PAH-contaminated Soil Remediation

Abstract Polycyclic aromatic hydrocarbons (PAHs) pose significant threats to soil and human health due to their toxicity, mutagenicity, and carcinogenicity. Consequently, there is a pressing need to develop efficient and environmentally friendly methods for cleaning up PAH-contaminated soil to protect the environment and human well-being. This study investigated the efficacy of a hybrid ultrasonication-ozonation technique for remediating soil contaminated with anthracene and phenanthrene. Various experiments were conducted to assess the removal efficiency of the hybrid remediation process, considering factors such as experiment duration, water volume, injected ozone, and ultrasonic power as independent variables. Utilizing a dataset comprising 150 data points, three machine learning algorithms were employed to establish the relationship between independent variables and contaminant removal efficiency. The eXtreme Gradient Boosting Regression (XGBR) model exhibited robust performance, achieving an R 2 score of 0.999 in the training set and over 0.83 in both testing and cross-validated sets. Analysis revealed that initial contaminant concentration, remediation process time, and ultrasonic power significantly influenced anthracene removal, while remediation process time, ozone concentration, and initial contaminant concentration were critical for phenanthrene removal efficiency. The XGBR model was further utilized to predict removal efficiency using an artificial dataset, and the results were visualized through four-dimensional plots, aiding in the optimization of parameters for soil remediation. This study underscores the potential of the hybrid ultrasonication-ozonation technique for PAH-contaminated soil remediation, highlighting the effectiveness of machine learning optimization in predicting and optimizing anthracene and phenanthrene removal efficiency across varying conditions.