Voltage Prediction of Proton Exchange Membrane Fuel Cells in Various Air Stoichiometries Using a Deep Learning Model Approach

In this paper, the ability of a predictive deep learning model that uses long short-term memory architecture to accurately predict the operating variables of a real operating proton exchange membrane fuel cell was explored. The model was built upon experimentally obtained output voltage data depending on the cathode air stoichiometry input variable. Raw data was preprocessed into a windowed dataset that is further used to build the prediction model. The model was first trained and then tested on the different stoichiometry values where great accuracy and generalization abilities were observed. Results give promising preliminary proof for a further upgrade with more data generated at various operating conditions, and to build a more sophisticated model that can predict fuel cell degradation based on limited diagnostic measurement data.