AI-DRIVEN MODELLING AND OPTIMIZATION OF DYNAMIC ELECTROCHEMICAL RESPONSES IN PROTON EXCHANGE MEMBRANE WATER ELECTROLYSIS SYSTEMS

Abstract

Background: Proton Exchange Membrane (PEM) water electrolysis is an important technology for sustainable hydrogen production, especially for integration into renewable energy systems. However, achieving the best dynamic electrochemical response in PEM is difficult due to the complicated interplays among different parameters for the operation, which involve temperature, pressure and current density. Objective: The objective of this study is to build AI-based models to predict and optimize the performance of PEM water electrolyzers for various operating points, leading to improved hydrogen yield as well as overall energy efficacy. Method: The operation data of ten PEM electrolysers (input parameters: temperature, pressure, and current density, performance indicators: hydrogen production rate, energy consumption and voltage efficiency) were collected. Diverse machine learning algorithms, such as Artificial Neural Networks (ANNs), Support Vector Machines (SVMs) and Decision Trees, were employed to develop the predictive models. Grid Search and Genetic Algorithms were used to optimize the hyper parameters of the models. PCA and SHapley Additive ex-Planations (SHAP) are used for feature selection. For operating conditions, a reinforcement learning and an evolutionary algorithm were used to tune system parameters while operating. Results: The accuracy of the ANN model was high (R² = 0.93). System efficiency was significantly increased by 30% hydrogen production and reduction of 15% energy consumption after optimization using reinforcement learning. Conclusion: The AI-based models considerably improve the performance and cost-effectiveness of PEM water electrolysis stacks. The findings underscore the power of machine learning and optimization methods for innovation in hydrogen generation technologies towards sustainable energy.