Data-Driven Temperature Deviation Estimation of a Proton Exchange Membrane Water Electrolyzer Stack from Electrochemical Impedance Using Machine Learning

We developed a method to analyze temperature variations in proton exchange membrane water electrolysis stacks by using machine learning to interpret electrochemical impedance spectroscopy (EIS) data. To train the machine learning model, we artificially generated impedance data at various temperature differences by merging data from uniformly heated stacks. The impedance data were collected using a dual active bridge DC/DC converter to apply a range of DC currents and an AC current through phase shift fluctuations. These data were then sampled for impedance analysis. A deep neural network was employed to correlate the real and imaginary components of impedance across frequencies with the stack's average temperatures and temperature variances. Our model, evaluated with actual thermal condition data, successfully predicted average temperatures and temperature differences within 2°C and 5°C accuracy, respectively, showing its potential for monitoring and managing temperature and other parameters in electrolysis stacks only by EIS and a deep neural network.