【摘要】 深度学习因其能够深入挖掘传感器采集的机器退化信息，在质子交换膜燃料电池(PEMFC)性能退化预测中取得了成功。然而，这些方法更侧重于时间相关性，而忽略了传感器的空间相关性。作者提出一种结合图卷积神经网络与长短期记忆网络(GCN-LSTM)的模型，该模型考虑了时间及空间相关性。首先，通过构建传感器网络融合多源数据。其次，运用GCN分析多源数据，提取空间特征。最后，通过LSTM学习时间特征，预测PEMFC性能退化。基于IEEE数据集验证了所提出的模型，拟合系数R~2在0.99以上。与仅考虑时间相关性的传统方法相比，均方根误差(RMSE)最大降低75.7%,预测精度得到了提升。更多还原

【Abstract】 Deep learning has been successful in predicting the performance degradation of Proton Exchange Membrane Fuel Cells(PEMFCs) by deeply mining the machine degradation information collected by sensors.However, these methods tend to emphasize temporal correlations while overlooking the spatial correlations of the sensors.This paper proposes a model that combines Graph Convolutional Neural Networks(GCN) with Long Short-Term Memory Networks(LSTM),referred to as GCN-LSTM,which considers both temporal and spatial correlations.Firstly, a sensor network is constructed to integrate multi-source data.Secondly, GCN is used to analyze multi-source data and extract spatial features.Finally, LSTM is employed to learn temporal features and predict PEMFC performance degradation.The proposed model is validated based on the IEEE dataset, achieving a fitting coefficient R~2 above 0.99.Compared to traditional methods that only consider temporal correlations, the Root Mean Square Error(RMSE) is reduced by up to 75.7%,indicating an improvement in prediction accuracy.更多还原