Performance evaluation and optimization of pharmaceutical removal with sulfate radical-based photooxidation processes by machine learning algorithms

Abstract

The aim of this study was to evaluate the efficiency of sulfate radical-based photooxidation using machine learning algorithms in the removal of metformin (METF), one of the most widely used pharmaceuticals in the world. UVC lamps were used in photochemical oxidation processes, and peroxydisulfate (PS) and peroxymonosulfate (PMS) were added as oxidants. The effects of UV-based process variables (initial pH, PS/PMS dose, initial METF concentration) on METF removal and the optimum conditions were determined. Under optimum conditions, the effect of inorganic anions, dominant radical species, and unit energy consumption (EE/O) was determined. The removal efficiencies of METF were 53.9 % and 58.3 % for the UV/PS and UV/PMS processes, respectively, under optimum conditions (4 mM PS dose and pH 7 for the UV/PS process; 8 mM PMS dose and pH 9 for the UV/PMS process). For both processes, nitrate decreased the METF removal rate while sulfate and phosphate were ineffective. The effect of bicarbonate and chloride was positive in the UV/PMS process and negative in the UV/PS process. Based on contribution rates, the dominant radical types were sulfate and hydroxyl radicals in the UV/PS and UV/PMS processes, respectively. EE/O values were determined as 1.19 and 1.05 kWh/L for the UV/PS and UV/PMS processes, respectively. METF removal was effectively modeled using machine learning algorithms, yielding high R² values and low MAE and RMSE levels. XGBoost models performed well, with no overfitting and successful generalization.