Abstract
Fuel cells have attracted widespread attention as power generation devices using clean energy. However, the lack of durability has limited further widespread commercialisation of fuel cells. Prediction of health indicators is a prerequisite and basis for improving durability of fuel cells. Nevertheless, the recoverable fault behavior in fuel cells remains one of the key challenges in prognostic. To address this issue, this paper proposes a prognostic strategy to evade recoverable fault by identifying them. The data volume is balanced using data augmentation techniques to detect recoverable fault based on classification results. The proposed prognostic strategy is validated on the aging dataset. The experimental results show that the accuracy of identifying the operational state increases to 97.11% after data enhancement. Prognostic during recoverable fault enables effective evasion, with an average decrease in root mean square error of 63.12%.

This publication has 8 references indexed in Scilit: