| In drinking water utilities, a strict control of residual chlorine levels is required during the treatment process and within the distribution system. Chlorine doses applied during water treatment are in many cases adjusted manually according to the information on residual chlorine measured downstream from field sampling or from on-line monitors. However, such information is available with a time delay which is associated to the travel time of water in the plant, in the clear water reservoirs and in the distribution system. In this work, an artificial neural network (ANN) modeling approach is proposed to predict the residual chlorine evolution in treated and distributed waters. The Quebec City water utility is studied in the research. Operational and water quality information used for the models was collected at five selected locations: at the post chlorination site, at the clear water reservoir and at the distribution system. A quality control program was developed to identify the non-representative data. Representative data were used to develop ANN models for predicting residual chlorine concentrations at a downstream location using operational and water quality parameter data from an upstream location. In total, six models were developed. Performances of these models were fairly good. Recommendations are offered to better qualify the data and improve the efficiency and the predictability of the models. |
