Feasibility-guided evolutionary optimization of pump station design and operation in water networks

Abstract

Pumping stations are critical elements of water distribution networks (WDNs), as they ensure the required pressure for supply but represent the highest energy consumption within these systems. In response to increasing water scarcity and the demand for more efficient operations, this study proposes a novel methodology to optimize both the design and operation of pumping stations. The approach combines Feasibility-Guided Evolutionary Algorithms (FGEAs) with a Feasibility Predictor Model (FPM), a machine learning-based classifier designed to identify feasible solutions and filter out infeasible ones before performing hydraulic simulations. This significantly reduces the computational burden. The methodology is validated through a real-scale case study using four FGEAs, each incorporating a different classification algorithm: Extreme Gradient Boosting, Random Forest, K-Nearest Neighbors, and Decision Tree. Results show that the number of objective function evaluations was reduced from 50,000 to fewer than 25,000. Additionally, The FGEAs based on Extreme Gradient Boosting and Random Forest outperformed the original algorithm in terms of objective value. These results confirm the effectiveness of integrating machine learning into evolutionary optimization for solving complex engineering problems and highlight the potential of this methodology to reduce operational costs while improving computational efficiency in WDNs.