Future prediction on the remaining useful life of proton exchange membrane fuel using temporal fusion transformer model

Proton exchange membrane fuel cell (PEMFC) is regarded as one of the leading developmental directions of the green power source. Predicting the remaining useful life (RUL) of online PEMFCs is challenging, primarily due to factors such as complex nonlinear degradation patterns and unknown data. To enhance the performance of RUL prediction, we present a deep learning model-based temporal fusion transformer model to predict the future total voltage of a PEMFC stack. The performance of RUL predictions was evaluated using different starting RUL prediction points to compare them with the existing study. The experiments are conducted with different future prediction steps for future RUL prediction to determine the model's prediction step limit. The proposed method achieved relative errors of 0.0148 %, 0.0060 %, 0.0662 %, and 0.0486 % for training lengths 300, 400, 500, and 600, respectively, in the FC1 and 0.0326 % in the FC2 datasets compared to the existing study scenarios. The experimental analysis resulted in relative errors of 0.0738 % for the FC1 dataset, 1.0239 % and 1.6512 % for the two FC2 dataset scenarios.