Augmenting RC-GRU with ARIMA-Generated Trends: A Data-Efficient Framework for PEMFC Degradation Prediction

This paper proposes a novel data-driven method for long-term health prediction of proton exchange membrane fuel cell (PEMFC). In specific, the autoregressive integrated moving average (ARIMA) method is exploited to learn the long-term linear trend of degradation. Relying on the learned trend, the residual convolutional gated recurrent unit (RC-GRU) network is introduced, for the first time, to learn the local nonlinearity of degradation with short-segment degradation data. By combining the merits of two methods, the proposed ARIMA-RC-GRU approach can predict the future degradation of PEMFC accurately with only small sampling data.