Data-Driven Modeling and Predictive Control of Proton Exchange Membrane Water Electrolyzers for Green Hydrogen Production

In the global transition to clean energy, hydrogen has emerged as a promising energy carrier due to its high energy density and zero emissions. Among the various production pathways, water electrolysis is a sustainable approach for generating hydrogen using electricity. Proton Exchange Membrane Water Electrolyzers (PEMWEs) are particularly suitable for renewable integration due to their fast dynamic response and compact design, providing a direct pathway to green hydrogen when powered by renewable electricity. Nevertheless, challenges in efficiency, durability, and cost remain critical barriers to large-scale deployment.

In this work, an investigation was done on the dynamic behavior of a PEMWE using experimental data from a lab-scale cyber-physical prototype. By varying flow rates, currents, and inlet temperatures, and measuring outputs such as outlet temperature, voltage, and hydrogen production, we generated a baseline dataset for model development. This dataset was used to train a Recurrent Neural Network (RNN), capable of capturing nonlinear, time- dependent process dynamics by improving on a neural network (NN) framework. This is due to RNN’s incorporating feedback loops, giving them memory of past states and enabling them to capture time-dependent system behaviors. A model predictive control (MPC) strategy was then implemented using the RNN model. Unlike PID controllers, which respond based on current and past errors, MPC predicts future system behavior over a moving time horizon and optimizes control actions subject to constraints. Embedding the RNN in the MPC framework allows for nonlinear, data-driven forecasting of system behavior. This RNN–MPC approach combines accurate modeling with robust control, providing a pathway toward more efficient and reliable PEMWE operation.