Simulation and optimization of multiple permeable reactive barriers (multi-PRBs) for acid mine drainage (AMD) based...

Multiple permeable reactive barriers (multi-PRBs) are an effective in-situ technology for acid mine drainage (AMD) treatment. However, their practical implementation is hindered by unclear mechanisms and a lack of decision models. In this study, a coupled processes numerical model was developed to simulate the synergistic removal of TFe and SO₄2⁻ through multi-PRBs with the optimized sequence of limestone, followed by biochar and then D201 resin. Machine learning integrated with the Non-dominated Sorting Genetic Algorithm (ML-NSGAII) was proposed for optimization, in which a Backpropagation Neural Network (BPNN) served as a highly accurate surrogate model (R² > 0.99) to predict system performance, reducing the computational load by 99.7% compared to conventional methods. Spearman correlation analysis and SHAP model interpretation revealed hydraulic load and filler size as the most influential parameters. Application of the TOPSIS-entropy weight method to the Pareto-optimal solution set yielded a final design that significantly enhanced system service life and treatment capacity while reducing costs. This research provides a practical and computationally efficient strategy for designing multi-PRBs for AMD treatment.