Research on Remaining Useful Life Prediction of PEMWE Stack Based on the LSTM-iTransformer Model.

The accurate prediction of the lifetime of the Proton exchange membrane water electrolyzer (PEMWE) stack is of critical importance for ensuring the stability of hydrogen energy storage systems. In order to address this issue, the present study has designed a series of experimental scenarios. These scenarios are intended to mimic the PEMWE stack's real-world performance in a hydrogen energy storage system, with a view to obtaining durability data. The data were then subjected to noise reduction using the S-G model, and the data were downscaled using the LightGBM model with the hydrogen generation rate as the performance degradation index of the PEMWE stack. In light of the findings, the LSTM-iTransformer model was employed to forecast the recession trend of the hydrogen generation rate on the steady state current data set. The validity of the model was subsequently verified on the dynamic current data set. The findings indicate that the mean errors of the model prediction under steady state and dynamic current conditions are 0.0372 and 0.0806 mL/min, respectively, demonstrating high prediction accuracy and validating the robustness and applicability of the model.