融合注意力机制的CNN-GRU燃料电池老化趋势预测OA北大核心CSTPCD

Accurate prediction of aging trend of fuel cells not only provides a reliable foundation for Prognostics Health Management and estimation of remaining useful life,but also plays an important role in enhancing safety.This paper proposes a CNN-GRU prediction model with attention mechanism.According to aging mechanism and Pearson correlation coefficient,an aging index comprising voltage,maximum voltage deviation rate and current is built as input.Then,the weight of CNN convolutional features is evaluated based on the attention mechanism to highlight important features and weaken minor features.Meanwhile,the impact of three attention mechanism modules (the squeeze and excitation module,efficient channel attention module and convolutional block attention module) on prediction performance is investigated.Our experimental results show the aging index proposed effectively enhances prediction accuracy.Compared with the baseline GRU model,the incorporation of the attention mechanism leads to a marked reduction in MAE and RMSE by at least 30.01％ and 29.39％.Notably,the CBAM-Block module performs the best with MAE and RMSE down by 72.72％ and 63.14％.