Akademik Veri Yönetim Sistemi
Fuel, cilt.410, 2026 (SCI-Expanded, Scopus)
Proton exchange membrane water electrolyzers (PEMWEs) are among the most promising technologies for sustainable hydrogen production. However, optimizing their operational performance remains a critical challenge. This study investigates the combined effects of anode and cathode electrocatalyst loadings on PEMWE performance using explainable machine learning (ML) approaches. A comprehensive experimental dataset of 1344 samples, incorporating parameters such as catalyst loadings (3-4 mgIrO2 cm−2 for anode and 0.4-0.7 mgPt/C cm−2 for cathode), membrane type (Nafion 115 and Aquivion E98-09S), temperature (50-80 °C), flow rate (50-100 mL min−1), torque (2-2.5 N·m), and current density were analyzed. Four ML models, Extreme Gradient Boosting (XGBoost), Random Forest (RF), Decision Tree (DT), and Categorical Boosting (CatBoost), were trained to predict current density under varying conditions. Bayesian hyperparameter optimization was applied to enhance predictive accuracy, with the DT model achieving the best performance (R2 = 0.9594), as validated by the Wilcoxon signed-rank test. SHapley Additive exPlanations (SHAP) analysis was used to interpret model outputs, identifying temperature and cathode catalyst loading as the most influential features. A nonlinear correlation was observed between catalyst loadings and current density. Best electrochemical performance was achieved with catalyst loadings of 0.6 mgPt/C cm−2 for platinum-carbon (Pt/C) composite at the cathode and optimized IrO2 loading at the anode. Furthermore, a cost-performance trade-off analysis revealed the most efficient configuration, offering a 14.8 % improvement in performance at reduced material cost. This study demonstrates the potential of explainable ML in guiding the design and optimization of PEMWEs, providing a data-driven framework for enhancing hydrogen production efficiency.