AbstractUrban water system optimization, e.g., chlorine dosage optimization, requires repeatedly running hydraulic and water quality models, which leads to significant computational and time costs for large-scale real networks. In the process of optimization, the search space of decision variables is adjusted downward to obtain the lowest-cost scheduling plans, which lead to a large number of negative samples and low optimization efficiency. To address this problem, this study proposes an efficient constraint-based pruning method (CBPM) that uses accumulated data during the optimization calculation to determine whether a sample meets the constraints before running simulation models, thereby pruning negative samples and improving optimization efficiency. An example network and a large-scale real network were used as case studies to demonstrate the performance of the proposed CBPM. The results show that the proposed CBPM significantly can improve optimization results using the same calculations as the simulation model. For example, application in a real water distribution network showed that to obtain the same total chlorine dosage solutions, the model using the proposed CBPM saved 57.9% of the calculations compared with the original optimization model.