The Degradation of Sulfamethoxazole via the Fe²⁺/Ultraviolet/Sodium Percarbonate Advanced Oxidation Process: Performance, Mechanism, and Back-Propagate–Artificial Neural Network Prediction Model

The degradation of sulfamethoxazole (SMX) via the Fe²⁺/Ultraviolet (UV)/sodium percarbonate (SPC) system was comprehensively investigated in this study, including the performance optimization, degradation mechanism, and predicting models. The degradation condition of SMX was optimized, and it was found that appropriate amounts of $C_{{\mathrm{Fe}}^{2+}}$ (10~30 μM) and C_SPC (10 μM) under an acidic condition (pH = 4~6) were in favor of a higher degradation rate. According to probe compound experiments, it was considerable that $\bullet \mathrm{OH}$ and $\bullet {\mathrm{CO}}_3^{-}$ was the primary and subordinate free radical in SMX degradation, and $k_{\bullet \mathrm{OH},\mathrm{SMX}}$ maintained two times more than that of $k_{{\bullet \mathrm{CO}}_3^{-},\mathrm{SMX}}$, especially under acidic conditions. The UV direct photolysis and other active intermediates were also responsible for the SMX degradation. These active intermediates were produced via the Fe²⁺/UV/SPC system, involving ${\bullet \mathrm{HO}}_2$, ${\mathrm{HCO}}_4^{-}$, $\bullet {\mathrm{O}}_{2 }^{-}$, or ¹O₂. Furthermore, when typical anions co-existed, the degradation of SMX was negatively influenced, owing to ${\mathrm{HCO}}_3^{-}$ and ${\mathrm{CO}}_3^{2-}$ possibly consuming $\bullet \mathrm{OH}$ or H₂O₂ to compete with SMX. In addition, the prediction model was successfully established via the back-propagate artificial neural network (BP-ANN) method. The degradation rate of SMX was well forecasted via the Back-Propagate–Artificial Neural Network (BP-ANN) model, which was expressed as ${\mathrm{Y}}_{\mathrm{pre}}=\tanh (\tanh ({\mathrm{x}}_{\mathrm{i}}{\mathrm{W}}^{\mathrm{ih}}){\mathrm{W}}^{\mathrm{ho}})$. The BP-ANN model reflected the relative importance of influence factors well, which was pH > t > $C_{{\mathrm{Fe}}^{2+}}\approx C_{\mathrm{SPC}}$. Compared to the response surface method Box–Behnken design (RSM-BBD) model (R² = 0.9765, relative error = 3.08%), the BP-ANN model showed higher prediction accuracy (R² = 0.9971) and lower error (1.17%) in SMX degradation via the Fe²⁺/UV/SPC system. These findings help us to understand, in-depth, the degradation mechanism of SMX; meanwhile, they are conducive to promoting the development of the Fe²⁺/UV/SPC system in SMX degradation, especially in some practical engineering cases.