Management of free residual chlorine concentrations in water distribution networks using deterministic and data-driven modeling techniques

Chlorine is the most common disinfectant for drinking water since it is cheap, effective, easy to apply and widely available. Chlorine reacts with both organic and inorganic compounds present in water and also with pipe wall. As a result of these reactions chlorine decay in water distribution networks can be classified under the name of chlorine bulk decay (Kb) and chlorine wall decay (Kw), respectively. As a result of absence of chlorine or very low chlorine concentrations in water distribution networks, the risk of waterborne diseases increases in case of water contamination. On the other hand, the presence of high chlorine concentrations in water distribution networks is associated with the formation of disinfection by-products and some of these disinfection by-products may cause cancer. Further, they may have other chronic and acute adverse health effects to human beings and animals. Consequently, chlorine concentrations should be kept within certain limits to minimize health risks. Water distribution networks are dynamic systems where hydraulic and water quality parameters show changes spatially and temporally. Therefore chlorine management can be achieved by only dynamic modeling studies. This study was conducted to manage chlorine dosing rates in Konyaalti water distribution network using deterministic and data-driven modeling techniques such as artificial neural network (ANN), auto regressive with exogenous input (ARX), auto regressive moving average with exogenous input (ARMAX) and process models. For this purpose, eight management scenarios that take into consideration extreme conditions in Konyaalti water distribution network were investigated using deterministic. The data sets required to set, calibrate and verify the deterministic and data-driven models were derived from the online continuous monitoring, sampling program and lab work. The study showed that data-driven modeling can be considered as a potential alternative to model chlorine concentrations in cases where the physical properties of water distribution networks, that enable deterministic modelling, are not available. The study revealed that online monitoring provides excellent data sets for chlorine modeling and management that ends up with automatic application of chlorine dosing. As a result of the study, the required residual chlorine concentration at the feeding station was determined as 0.40 mg/L in winter season while it was determined as 0.50 mg/L in summer season and 0.40 – 0.50 mg/L in the other seasons.