The insufficient durability of proton exchange membrane fuel cell (PEMFC) has remained one of the challenges hindering their large-scale commercialization. This paper proposed a method utilizing Bayesian optimization (BO) algorithm was to optimize the echo state network (ESN) model for predicting the degradation of PEMFC performance. BO was employed to obtain the optimal hyperparameters for the ESN model

which was then used to predict the voltage of PEMFC. Additionally

a decline in voltage serves as a significant indicator of PEMFC performance degradation

with areas of rapid voltage decline containing more information on performance degradation features

necessitating more frequent sampling. Conversely

areas with minor voltage declines contain less information on performance degradation features

requiring lower frequency sampling. Thus

this paper introduces an adaptive fuzzy sampling (AFS) method for dataset sampling to enhance the prediction accuracy of PEMFC performance. Experimental results show that

in static conditions

the BO-ESN model reduces the root mean square error (RMSE) and mean absolute percentage error (MAPE) by 52.4% and 63.6%

respectively

compared to the ESN model. After applying AFS sampling

the RMSE and MAPE of the BO-ESN model are further reduced by 49.8% and 54.5%

respectively

compared to fixed interval sampling. In dynamic conditions

the RMSE and MAPE of the BO-ESN model decrease by 13.4% and 7.96%

respectively

compared to the ESN model. According to the experimental results

this method demonstrates superior performance in predicting the performance degradation of PEMFC.