Predicting disinfection by-products (DBPs) in supply water within a real water distribution network using an...

This study develops an artificial neural network (ANN) model using advanced learning algorithm techniques to predict disinfection by-products (DBPs) in supply water. The model is based on real-scale data from the Chattogram water distribution network (WDN), a coastal city in Bangladesh. The research revealed a significant presence of free halide components in both the source and distributed water, indicating the potential formation of bromo- and iodo-DBPs.The developed model incorporates nine water quality parameters, including pipeline distance (instead of contact time used in existing models) and free iodine (I^-) as two new parameters. The radial basis function (RBF) algorithm was employed to predict the total trihalomethanes (TTHMs), which are the dominant DBPs, along with their four major species in actual water distribution networks. The model was trained and validated with 120 data points to assess its prediction accuracy. The RBF-ANN model demonstrated good predictability, with regression coefficients (R²) ranging from 0.81 to 0.87. The prediction errors, with a maximum of twenty neurons (MN) and sixty spreads (S), showed that this arrangement enhances predictability. A 10-fold cross-validation approach was adopted to ensure the model's robustness and generalization, with consistent performance across all folds. Furthermore, the findings indicate that the prediction capability of the model improves as the volume of data increases. Performance analysis revealed that using all nine input parameters together resulted in the most accurate predictions of TTHM concentrations. The developed RBF-ANN model may, therefore, be effective for predicting the studied DBPs in actual water supply systems, particularly in halide-rich water.