Maintaining water quality parameters in water distribution systems (WDS) is essential to prevent spread of waterborne diseases, making it an essential goal for water utilities. Chlorine is usually used as a disinfectant because of its ease of application and low cost. However, in excess can lead to the formation disinfection by-products (DBPs) as also produce smell and taste in water [1]. The concentration of the disinfectant is influenced by the dosing rate as also by the hydraulics of the system [2]. Due to travel time of water between the dosing and measurement points, the chlorine concentration can decrease as it travels through the distribution system, having the so-called chlorine decay [2-3]. Hydraulic simulators are typically used to model these systems, although these require a good understanding of the system and a large time to calibrate physical characteristics. Additionally, each application needs to determine a chlorine decay coefficient, which complicates replicating the model for different WDS [3]. Alternatively, data-driven models can learn chlorine kinetics from measured data and empirical relationships between several dependent and independent variables [2]. Previous research has analysed the use of data-driven techniques in simulating chlorine decay, although these studies have exclusively focused on feedforward architectures [3]. This work aims to use machine learning models for the space and time-series prediction of chlorine residuals in a real WDS. The use of recurrent algorithms aims to mitigate the time-delay effect on the spread of the chlorine concentration through the WDS, aiming to predict residual chlorine levels at distribution points, mainly in those far away from the injection station where the time travel of water is larger. Integrating these predictive models with an optimization framework, this work enables an optimal injection dosing operation, reducing chlorine usage and associated costs, while maintaining water quality across the entire WDS. Together, these components are part of a broader smart predictive digital twin, which enhances water quality management through efficient and adaptive chlorine injection strategies.