An ensemble learning method to predict a hydrogen production rate in the viable wastewater-hydrogen supply chain network design

Hydrogen production from industrial wastewater has attracted attention as a promising complementary pathway toward sustainable energy. This study investigates dark fermentation by considering key factors, such as biomass concentration, Ni, pH, Fe, hydraulic retention time, butyrate, acetate, the acetate-to-butyrate ratio, chemical oxygen demand, and ethanol. Four machine learning models, bootstrap aggregating (Bagging), random forest, gradient boosting regression, and XG-Boost regression, were applied to predict hydrogen yields. Among them, XG-Boost regression performed best, achieving an overall R² of 0.966 and MSE of 0.111. Based on these predictions, a two-objective optimization model was developed to design a sustainable and resilient hydrogen supply chain network, targeting profit maximization and non-resiliency minimization. Using the ε-constraint method and validated through an illustrative application in Iran, showing improved profitability and reduced operational risks, with sensitivity analyses confirming the dominant impact of transportation on CO₂ emissions and responsiveness on computational complexity. The novelty of this work lies in being the first to integrate wastewater-derived hydrogen production data, advanced machine learning prediction, and multi-objective supply chain optimization into a unified framework, bridging the gap between bio-hydrogen potential assessment and strategic supply chain design.