Durability estimation and short-term voltage degradation forecasting of vehicle PEMFC system: Development and evaluation of machine learning models. (15th November 2022)

• Record Type: Journal Article
• Authors: Liu, Ze; Xu, Sichuan; Zhao, Honghui; Wang, Yupeng
• Is Part Of: Applied energy. Volume 326(2022)
• Publication Date: 2022-11-15
• Subjects: Vehicle PEMFC system -- Degradation forecasting -- Durability estimation -- Data-driven modeling -- Machine learning; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2022.119975
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: The periodic durability dataset of the vehicle fuel cell system is extracted for the development of degradation modeling and prediction technology. The prediction performance of the developed LSTM model, SVR model and LSTM-SVR combined model in the short-term prediction stage is comparatively evaluated. An innovative framework based on LSTM model rolling deployment and prediction is proposed to achieve quick and accurate estimation of long-term durability. Abstract: Proton exchange membrane fuel cell (PEMFC) systems are emerging as one of the most promising solutions for carbon neutrality in transportation, however, durability problem remain a major obstacle to their large-scale commercialization. Developing an accurate model to predict the short-term aging state and long-term durability level of PEMFC is conducive to formulating optimal measures in time and further to improving durability. In this paper, the short-term voltage degradation and long-term durability level are predicted and evaluated by combining machine learning (ML) methods with time-series voltage degradation data obtained under vehicle dynamic load. In the short-term forecasting stage, the long short-term memory (LSTM) model, the support vector regression (SVR) model, and the LSTM-SVR combination model are developed respectively, and the prediction results of the three models are compared and evaluated. The LSTM-SVR combined model achieved the best short-term prediction accuracy of 96.6%, followed by LSTM model (95.5%). Considering the difficulty of model deployment and the feasibility of quickly assessing long-term durability in practical application, a LSTM-based model rolling prediction mechanism is proposed to rapidly evaluate the long-term durability index of the developed PEMFC system, the results show that the proposed forecasting model and method can accurately predict the voltage degradation trend and quickly evaluate the long-term durability level, which not only makes contributions to greatly saving the durability R&D costs, but also provides the possibility to adjust the optimization measures in real-time to further improve the durability according to the prediction results.