ResBlock-BiGRU-IAT: A hybrid model integrating residual convolution and improved attention mechanism for accurate prognostics in PEMFC

Proton exchange membrane fuel cells (PEMFCs) are widely used in transportation, portable power, and stationary power generation due to their high efficiency and environmentally friendly by-products. However, the insufficient lifetime evaluation hinders their development. In this paper, bidirectional gated recurrent units (BiGRU), BiGRU with convolutional neural network (CNN-BiGRU), and BiGRU with attention mechanism (BiGRU-AT) were developed. The prediction performance of the three models was compared using training sets with two different sequence lengths on both static and quasi-dynamic datasets. The impact of incorporating convolutional operation and attention mechanism on BiGRU prediction performance was analyzed in different scenarios. Finally, based on the analysis results, a BiGRU with residual block and improved attention mechanism (ResBlock-BiGRU-IAT) model was proposed. Compared to the previous three models, this model demonstrated significant advantages in prediction performance in different scenarios. Remaining useful life (RUL) was predicted through the ResBlock-BiGRU-IAT model, and the RUL prediction accuracy was evaluated using absolute error (AE). The results show that the attention mechanism enhanced the model's performance in long sequence prediction tasks, and the convolutional operation improved the model's adaptability. The AE for RUL prediction of 0.25 ~ 0.56 h and 0.45 ~ 0.65 h, respectively.