341 - Black-box Modeling of Disinfectants and Disinfectant By-products in Water Distribution Systems

Water distribution networks (WDNs) are dynamic systems that adapt according to the quantity and quality of water required to supply the population. The complexity of hydraulic and water quality modelling of a network is increased by the number of variables in space and time that must be included to analyze the behavior of the system. Field instrumentation has allowed the collection of data and generation of historical information to provide input for the implementation of predictive performance models. Regarding water quality, artificial intelligence modeling techniques have contributed to reduce the uncertainty and computational time of complex models related to applications such as intentional and unintentional events, dispersion of contaminants, modelling in real-time, etc. In this study, machine learning techniques are explored to model water quality in a distribution system. Black box modelling is used to assess the accuracy of predicting residual chlorine and the formation of disinfection by-products (DBPs) in WDSs. Also, the sensitivity to the number of parameters in the inputs and the generation of different initial scenarios. For this work, the case study is the trunk network of Bogotá, Colombia, which, due to its topology and size, has been used in previous research. The purpose of this research is to compare and evaluate machine learning models that have not been widely explored in the past. The results will be assessed with different evaluation metrics for comparison and discussion in order to identify the limitations of the models and future research directions.