Utilizing machine learning for reactive material selection and width design in permeable reactive barrier (PRB)

Area of Science:

• Environmental Engineering
• Water Treatment Technologies
• Machine Learning Applications

Background:

• Permeable reactive barriers (PRBs) are crucial for groundwater remediation.
• Optimal material selection and width estimation pose significant challenges in PRB design.
• Machine learning (ML) offers potential for predicting contaminant transport and evolution.

Purpose of the Study:

• To develop and validate an ML model for designing PRBs.
• To improve the accuracy of predicting PRB performance and dimensions.
• To enhance the efficiency of PRB design through data-driven approaches.

Main Methods:

• Development of an ML algorithm for PRB design.
• Validation using experimental data and a case study.
• Application of SHapley Additive exPlanation (SHAP) for factor analysis.
• Integration of an evaluation index with expert opinions for material selection.
• ML-based prediction of mass transport zone width.

Main Results:

• ML models demonstrated high accuracy in predicting Freundlich equilibrium parameters (R² 0.94-0.96).
• Refining input parameters based on SHAP analysis further improved prediction accuracy (R² 0.99).
• ML accurately predicted mass transport zone width (R² 0.98, RMSE 1.2).
• ML-based design enhanced efficiency and reduced experimental time compared to traditional methods.

Conclusions:

• ML provides a feasible and accurate approach for PRB design, optimizing material selection and width estimation.
• The developed ML model significantly improves design efficiency and reduces experimental workload.
• Further data expansion and algorithm optimization hold potential for broader applications of ML in PRB technology.