<p>Proton Exchange Membrane Fuel Cells (PEMFCs) are gaining increasing prominence in the energy sector, and the prediction of their Remaining Useful Life (RUL) has attracted significant attention. Based on the time-varying trend of PEMFC voltage signals, this paper proposes a WHT-VMD-IASO-BiLSTM model for PEMFC RUL prediction. Firstly, Wavelet Hard Thresholding (WHT) is adopted to denoise the original voltage signals. Subsequently, Variational Mode Decomposition (VMD) is used to decompose the denoised data into multiple Intrinsic Mode Functions (IMFs), and the IMF signals are selected based on the Fourier spectrum to reconstruct the voltage signals. Then, the Tent chaos mapping is utilized to initialize the population of the Atomic Search Optimization (ASO) algorithm, the number of high-quality atoms is dynamically adjusted, and an exponential decay strategy is introduced to optimize the velocity update of ASO. The optimized Improved ASO (IASO) is applied to optimize the weights and biases of the Bidirectional Long Short-Term Memory Network (BiLSTM). Finally, the proposed model is used for training and prediction based on voltage signals, and experimental comparisons are conducted with existing models. The results demonstrate that the proposed model achieves higher prediction accuracy.</p>