联合G-WD与优化TCN-Transformer混合特征输入的PEMFC老化趋势预测OA

The aging trend prediction of PEMFCs(proton exchange membrane fuel cells)isimportant for predicting the lifespan of PEMFCs.To improve the prediction accuracy in the voltage aging trend of PEMFC stacks,this paper proposes a hybrid feature-input aging trend prediction model that combines G-WD and optimized TCN-Transformer.The original stack voltage data were processed through data down-sampling and smoothing.Correlation analysis was performed to identify strongly correlated physical parameters as input features.After decomposition and smoothing processing with G-WD,the voltage data is used as strong feature input,and time-domain feature extraction is performed as auxiliary feature input.In the TCN-Transformer model,an attention mechanism is inserted into the TCN block,constructing the TCN-SE-Transformer model.The input features are fed into the model for training,and the DE(differential evolution)algorithm is used to optimize different hyperparameters of the TCN-Transformer.The DE-TCN-SE-Transformer model's prediction results are obtained when the prediction error is minimized.Compared with other models and conditions,this model achieves higher prediction accuracy in the aging prediction of PEMFC stack voltage.