Explainable machine learning modeling and multi-objective optimization of cation exchange resin degradation via supercritical water oxidation

To achieve synergistic optimization of external specific energy consumption at laboratory scale ( E lab,ext ) and COD removal efficiency ( R COD ) during supercritical water oxidation of cation exchange resins, this paper constructs an interpretable machine learning model and a multi-objective optimization algorithm. Give the small sample size (n=58), leave-one-out cross-validation (LOOCV) was used to screen the prediction model. The results show that Gaussian process regression (GPR) performed best (MAE=1.057%, RMSE=1.359%, R ²=0.884). SHAP analysis showed that temperature and oxidant stoichiometry were the main controlling factors affecting R COD , and residence time played a crucial role in COD deep degradation. Furthermore, the temperature effect exhibited significant nonlinearity and diminishing marginal returns. Based on this, a GPR-NSGA-II optimization model was constructed at the laboratory scale, aiming to maximize R COD and minimize E lab,ext . The recommended operating conditions were determined using TOPSIS: T = 470.9 ℃, O 2 sto = 135%, t = 3.25 min, COD i = 8 × 10⁴ mg·L⁻¹, w NaOH = 1.97 wt%. This corresponds to an R COD of 99.51% and an E lab,ext of 40981 kJ·kg⁻¹ COD (11.38 kWh·kg⁻¹ COD). These findings offer robust methodological support for process optimization and engineering decision-making in the SCWO treatment of high-concentration organic wastes. • An interpretable machine learning framework for SCWO treatment of cation exchange resins was constructed. • SHAP analysis revealed that temperature and oxidant stoichiometry are the primary parameters affecting COD removal efficiency. • The GPR-NSGA-II algorithm achieves a synergistic optimization of 99.51% R COD and low energy consumption at the laboratory scale. • The study demonstrated the risks of salt deposition and corrosion faced by the optimal operating conditions at the laboratory scale when applied to the industrial scale.