Lifetime Prediction Analysis of Proton Exchange Membrane Fuel Cells Based on Empirical Mode Decomposition—Temporal Convolutional Network

Proton exchange membrane fuel cells (PEMFCs) are ideal for fuel cell vehicles due to their high specific power, rapid start-up, and low operating temperatures. However, their limited lifespan presents a challenge for large-scale deployment. Accurate assessment of remaining useful life (RUL) is essential for enhancing longevity. Automotive PEMFC systems are complex and nonlinear, making lifespan prediction difficult. Recent studies suggest deep learning approaches hold promise for this task. This study proposes a novel EMD-TCN-GN algorithm, which, for the first time, integrates empirical mode decomposition (EMD), temporal convolutional network (TCN), and group normalization (GN) by using EMD to adaptively decompose non-stationary signals (such as voltage fluctuations), the dilated convolution of TCN to capture long-term dependencies, and combining GN to group-calibrate intrinsic mode function (IMF) features to solve the problems of modal aliasing and training instability. Parametric analysis shows optimal accuracy with the grouping parameter set to 4. Experimental validation, with a voltage lifetime threshold at 96% (3.228 V), shows the predicted degradation closely aligns with actual results. The model predicts voltage threshold times at 809 h and 876 h, compared to actual values of 807 h and 872 h, with a temporal prediction error margin of 0.250–0.460%. These results demonstrate the model’s high prediction fidelity and support proactive health management of PEMFC systems.