A Holt Trend-Enabled Attention-ConV-GRU Method for Long-Term Degradation Prediction of Fuel Cells

Abstract Proton Exchange Membrane Fuel Cell (PEMFC) has attracted widespread attention as a promising energy conversion device. However, the lack of durability is one of the major reasons limiting the further development of PEMFC. Accurately predicting the Remaining Usable Life (RUL) of a PEMFC is a key prerequisite for durability enhancement. This paper is dedicated to addressing the limitations of the training data interval for RUL prediction performance. Based on the above ideas, we propose a Holt trend-enabled Attention-Convolution-Gated Recurrent Unit method for extracting trend information and making multi-step predictions. The experimental results show that this strategy improves the multi-step prediction and maintains the inherent degradation trend over a longer range of time steps. The Root Mean Square Error and Mean Absolute Error both fell by more than 45%. The prediction accuracy reach 93.76%.