Work

Multi-objective optimization of water hammer protection for long-distance water conveyance systems based on hybrid stacking ensemble learning and mechanistic interpretation

Year: 2026

Type: article

Abstract: ABSTRACT Four-stage research framework diagram: hydraulic simulation data acquisition, BPNN-XGBoost stacking ensemble modeling, SHAP-based interpretability analysis, and multi-objective optimization using Pareto and CRITIC-TOPSIS for decision-making. Long-distance water conveyance systems are highly susceptible to destructive column separation water hammer during pump failures. Existing designs struggle with time-consuming calculations, insufficient prediction accuracy for extreme pressures, and unclear parameter mechanisms. This study proposes an optimization method integrating Stacking ensemble learning, interpretable mechanistic analysis, and multi-objective decision-making. A Stacking surrogate model based on back propagation neural network and extreme gradient boosting was constructed using adaptive weighted fusion. Test results demonstrate the model's superiority in predicting minimum water hammer pressure, with R2 increasing to 0.968 and root mean squared error reducing by 17.9%. Based on this, SHAP analysis revealed the dominant protective role of one-way surge towers and identified a critical threshold for the air valve outflow diameter where the ‘air cushion’ effect fails. Finally, a multi-objective optimization model combining NSGA-II with CRITIC-TOPSIS was established. The optimized scheme eliminated negative pressure risks while reducing the one-way surge tower volume by 22.8%. This study achieves an effective combination of high-precision prediction and mechanistic interpretation, offering a solution that balances safety and economic efficiency for large-scale water conveyance projects.

Source: AQUA - Water Infrastructure Ecosystems and Society

DOI: 10.2166/aqua.2026.004