Degradation prediction of 65 kW proton exchange membrane fuel cells on city buses using a hybrid approach with the advantage actor-critic method

Abstract

Degradation prediction of proton exchange membrane fuel cell (PEMFC) is critical for optimizing fuel cell operation and extending its lifetime, facilitating its large-scale commercialization. This paper proposed a hybrid method based on the mechanism voltage model and the advantage actor-critic (A2C) algorithm to achieve degradation prediction. A semi-empirical voltage model is established to extract four aging parameters: exchange current density, leakage current density, reaction area, and area-specific resistance. A2C is then applied to predict aging parameters and output voltage under stochastic conditions, precisely characterizing the degradation process. Two sets of ∼1,000h real data from 65 kW PEMFCs on city buses are used for training, validating, and optimizing the proposed model. The results show that the method can accurately predict voltage degradation and efficiently track voltage fluctuations without delay under actual operating conditions. The prediction root mean square errors for the two datasets with a training ratio 0.8 are 0.0095 and 0.0115, respectively, smaller than conventional prediction methods. The hybrid method can provide detailed internal degradation information and has potential for online forecast applications.