Optimization Techniques for Chlorine Dosage Scheduling in Water Distribution Networks: A Comparative Analysis

A sufficient dose of disinfectant needs to be applied to maintain a minimum residual throughout water distribution networks (WDNs). Yet, excessive dosing of chlorine-based disinfectants leads to the formation of hazardous disinfection byproducts. Several frameworks have been proposed in previous literature to optimize chlorine dosing schedules by minimizing the total dose while maintaining evenly distributed residuals throughout the WDN. Many of these studies relied on evolutionary algorithms (EAs), such as genetic algorithm (GA) and particle swarm optimization (PSO), which are known to require numerous evaluations of computationally expensive water quality (WQ) models. Additionally, EAs typically feature many parameters that require careful tuning. Recently, Bayesian optimization (BO) has been proposed as an alternative to EAs for the optimization of hydraulics and water quality in WDNs. To speed-up convergence, BO builds a probabilistic surrogate model (e.g., Gaussian Process) in place of the original black-box model, and then leverages the predictions to explicitly control the exploration/exploitation trade-off. Yet, it’s still unclear how BO’s performance compares against other popular WQ optimization techniques. This study aims to fill this knowledge gap by conducting a systematic comparison between the performance of BO, GA, and PSO. To that end, a comprehensive sensitivity analysis is conducted on each optimization approach to understand how different optimization parameters influence their performance. Preliminary results revealed that BO requires significantly fewer evaluations than GA and PSO to converge to high-quality solutions, while PSO displayed lower sensitivity to the change in parameters compared to BO and GA.