Reinforcement learning-based DSS for coagulant and disinfectant dosage selection on drinking water treatment plants

Abstract

Treatments to be applied for water purification must be dynamically adaptable to raw water conditions. Currently, treatments are applied based on standards that are not optimized for the circumstances of each drinking water treatment plant (DWTP), nor for critical events. This paper presents a methodology for the creation of an artificial intelligence (AI) DSS (decision support system), encompassing the principal steps of the drinking water treatment processes (coagulation, sedimentation, filtration, and disinfection), based on reinforcement learning techniques, that provide suggestions about the most efficient treatments (coagulant and chlorine dosages) for various raw water conditions, including critical events such as heavy rain and saline intrusions. Together with the model, a retraining strategy is included; so the DSS adapts itself to the specific circumstances of each different DWTP. The model has been developed and validated in a DWTP replica. Furthermore, the model has been provided to a real DWTP to obtain feedback from experienced staff. The results and evaluation of the model are promising as a first approach to a DSS for drinking water treatment suggestions, although future versions might require more water quality parameters to characterize raw water.