AEM Water Electrolyzer Performance Prediction and Multi-Objective Optimization Using a LightGBM-Based Surrogate Model

Abstract

This study proposes a LightGBM-based surrogate model for fast performance prediction and multi-objective optimization of an anion exchange membrane (AEM) water electrolyzer. A dataset of 5000 samples was generated from an electrochemical simulation model by varying key operating and electrochemical parameters, including current density, temperature, pressure, membrane thickness, exchange current densities, and saturation factor. The developed model was trained to predict three main performance indicators: applied voltage, hydrogen production rate, and electrical efficiency. The results showed high predictive accuracy for all outputs, with R 2 values of 0.94001 for applied voltage, 0.99983 for hydrogen production rate, and 0.94967 for electrical efficiency. The parity plots confirmed a strong agreement between the predicted and reference values. In addition, the analysis of electrolyzer behavior and the correlation study revealed a clear trade-off between hydrogen production, voltage demand, and efficiency. A Pareto-based multi-objective optimization was then performed to identify suitable operating conditions. The obtained Pareto-optimal set included representative points corresponding to maximum efficiency, maximum hydrogen production, and a best compromise solution. Overall, the proposed approach provides a fast and reliable tool for predicting and optimizing AEM water electrolyzer performance while reducing the computational cost of repeated electrochemical simulations.