Regression model and artificial neural network model to predict halonitromethane formation from amino acids during...

Halonitromethanes (HNMs) were high-toxicity nitrogenous disinfection byproducts generated by amino acids (AAs) during UV/monochloramine (UV/NH₂Cl) disinfection in bromide-containing water. HNM concentrations fell over time, highlighting disinfection time optimization as an effective control strategy. Using specific ultraviolet absorbance at 254 nm characterizing aromatic carbon composition, a single equation predicted HNM concentrations generated by three AAs (including tyrosine, tryptophan and aspartic acid) in bromide-containing real water and distilled water combined with NH₂Cl dose, disinfection time, Br^-, HCO₃^- and pH. Utilizing a second-order polynomial for disinfection time whose coefficients incorporated NH₂Cl dose and water quality parameters, regression model successfully described HNM concentrations, with improved accuracy by introducing nitrate. Due to excellent non-linear fitting abilities, back propagation (BP) and radial basis function (RBF) artificial neural network (ANN) models were applied, followed by principal components analysis (PCA) for data preprocessing and genetic algorithm (GA) for model structure optimization. The predictive performance in descending order was RBF ANN, regression model and BP ANN, with GA optimization outperforming PCA optimization. GA-RBF ANN model achieved R² = 0.90, mean absolute error = 0.39, mean relative error = 20% and root-mean-square error = 0.68, demonstrating better generalization and 12% fewer hidden layer neurons than RBF ANN model. It was further simplified by removing insignificant pH and HCO₃^-. HNM-associated cytotoxicity/genotoxicity based on the additivity of toxicity, respectively damaging cells/DNA, had a strong positive linear correlation (R² > 0.9) with its concentrations. It confirmed HNM concentration control as effective toxicity management. Heatmaps of HNM generation in bromide-containing real water showed that both low NH₂Cl doses (e.g., <6 × 10^-5 M) and high NH₂Cl doses (e.g., (1.5-2.5) × 10^-4 M) combined with high UV doses (e.g., >1684.8 mJ⋅cm^-2) could reduce HNM generation.