AI-DRIVEN MODELING AND OPTIMIZATION OF DYNAMIC ELECTROCHEMICAL RESPONSES IN PEM WATER ELECTROLYSIS SYSTEMS

Abstract

Proton Exchange Membrane (PEM) water electrolysis is key for sustainable hydrogen production, but optimizing dynamic electrochemical responses is challenging due to complex interactions among temperature, pressure, and current density. This study develops AI-driven models to predict and optimize PEM electrolyzer performance across various operating conditions to enhance hydrogen yield and energy efficiency. Data from ten PEM electrolysers were used to train models with machine learning techniques including Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Decision Trees. Hyperparameters were optimized via Grid Search and Genetic Algorithms, while PCA and SHAP were applied for feature selection. Reinforcement learning and evolutionary algorithms tuned operational parameters. The ANN model achieved high accuracy (R² = 0.93), and optimization improved hydrogen production by 30% and reduced energy consumption by 15%. These results demonstrate that AI-based modeling and optimization significantly boost PEM electrolyzer performance, advancing sustainable hydrogen generation