Kinetic Signature Decomposition and SHAP-Based Interpretability for Pretreatment Selection in Sludge Anaerobic Digestion

Anaerobic digestion (AD) of sludge is often limited by hydrolysis, and various pretreatments are deployed to accelerate this step. However, comparing pretreatments with incommensurable operating parameters remains challenging. Here, three mechanistically distinct pretreatments (zerovalent iron–peroxydisulfate (ZVI-PDS) oxidation, ultrasonication, and NaOH alkali) were systematically evaluated under mesophilic batch AD. All cumulative methane curves were fitted to the modified Gompertz model, projecting each condition onto three kinetic signatures (potential P, maximum rate Rmax, and lag phase λ). A unified feature space enabled predictive modeling and SHAP-based decomposition of which drivers control each kinetic dimension, with bootstrap resampling quantifying uncertainty. Results showed that ultrasonication gave the highest mean methane yield (97.24 mL/g VSadded) but the poorest cost-effectiveness and largest variability; ZVI-PDS offered the best cost-effectiveness (0.648 mL/g VSadded per CNY/m3) with narrow uncertainty; alkali was cheapest but was associated with possible sodium and alkalinity stress, showing the longest lag (λ̅ = 2.17 d). SHAP revealed that solubilization (ΔSCOD) dominated P and cumulative yield, whereas Rmax was cocontrolled by ΔSCOD and Fe0 dose, and λ bore a strong alkali-specific chemical penalty beyond solubilization. This kinetic fingerprint framework bridges heterogeneous pretreatments, exposes method-specific inhibition windows, and supports rational, uncertainty-aware pretreatment selection.