Abstract
Due to multiple coupled monitoring parameters existing in fuel cell (FC) systems, effectively identifying key features and capturing overall degradation trends have become extremely difficult. To solve these problems, a hierarchical stochastic gradient (HSG) algorithm and principal component analysis (PCA) based Hammerstein model is proposed to predict FC degradation. First, a Hammerstein prediction model with a nonlinear static polynomial module and a dynamic linear Controlled Auto Regression and Moving Average (CARMA) module is explored to model FC degradation. Second, by the PCA technique, the main characteristics which have significant influence on the performance of PEMFC are selected from many variables, and the key characteristics are accurately extracted. Third, by the hierarchical estimation technique, the Hammerstein model is recasted into two submodels to enable separated parameters being interactively updated through minimizing three defined cost functions and the HSG algorithm with high efficiency, so as to predict FC degradation. It has the advantages of fewer model parameters, fast convergence speed and high prediction accuracy. Finally, simulation experiments are carried out to evaluate model performance using root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R 2 ), and the PCA based Hammerstein model is verified to achieve high precision degradation prediction.
Funding Information
  • The National Natural Science Foundation of China (62273190)

This publication has 37 references indexed in Scilit: