High-Precision Prognostics of PEMFC Remaining Useful Life via VMD-IASO-BiLSTM

Proton exchange membrane fuel cells (PEMFCs) are gaining prominence in the energy sector because of their high efficiency and environmental benefits. However, predicting their remaining useful life (RUL) remains challenging due to complex electrochemical processes and sensitivity to operating conditions. This study proposes a novel VMD-IASO-BiLSTM model to accurately predict the PEMFC RUL based on the trends of the voltage signal. Firstly, the model integrates wavelet hard thresholding (WHT) for denoising raw voltage signals, followed by variational mode decomposition (VMD) to decompose signals into intrinsic mode functions (IMFs). Key IMFs are selected on the basis of their conformity to normal distributions for signal reconstruction. Then, the tent chaotic map initializes the atomic search optimizer (ASO) algorithm, which is enhanced by Gaussian mutation and a hyperbolic tangent function to optimize atom positions and velocities. The optimized improved atomic search optimizer (IASO) algorithm fine-tunes the weights and biases of a bidirectional long-short term memory (BiLSTM). Finally, experimental results show that the VMD-IASO-BiLSTM model achieves 99.98% prediction accuracy, significantly outperforming existing methods.