Computational Model Reduction in Water Distribution Networks for Enhanced Operational Optimization

<p dir="ltr">Water distribution networks (WDNs) are critical infrastructure that require efficient operational strategies to ensure reliability and minimize costs. However, the complexity of hydraulic models—often involving thousands of components and non-linear equations—limits their use in real-time optimization. This study addresses this challenge by combining computational model reduction with a genetic algorithm for energy-efficient optimization of pumping stations. The proposed method simplifies network complexity while preserving key hydraulic properties, resulting in faster and more effective optimization. The reduction process relies on skeletonization techniques based on graph theory. It systematically removes hydraulically insignificant pipes and nodes while maintaining connectivity and performance. Graph-based approaches decompose the network into subgraphs, each consolidated into a super node. For every block, set point curves are derived by simulating responses to varying inflow and demand conditions. These curves define the equivalent pressure-flow relationship at entry points, accurately replicating original hydraulic behavior. Set point curves capture the hydraulic behavior of simplified blocks. Together, these techniques reduce the optimization decision space while preserving essential hydraulic metrics like pressure and flow. The method was applied to the Murcia (Spain) WDN, reducing the model from 2368 nodes and 2600 pipes to 294 nodes and 365 pipes. All nine pumping stations were retained for optimization. The genetic algorithm ran more generations on the reduced model, with a 96% decrease in computation time per generation and a significantly lower average energy cost.</p><p dir="ltr">This paper was presented at the 21st Computing and Control in the Water Industry Conference (CCWI 2025) at the University of Sheffield (1st - 3rd September 2025).</p>