Volume 53

Degradation Prediction Model based on CEEMDAN-LSTM Hybrid Method Considering Reversible Degradation of Proton Exchange Membrane Fuel Cell Huijin Guo, Julong Zhou, Jinghui Zhao, Beiming Huang, Ruitao Li, Tiancai Ma

Abstract

The demonstration of proton exchange membrane fuel cell (PEMFC) technology for commercial vehicles has shown increasing success, but comprehensive commercialization remains limited by the service life of the cells. Timely prediction of fuel cell lifetime can enhance state assessment in advance, enabling precise control for performance recovery operations, thus reducing use and maintenance costs and avoiding predictable risks. Commonly used lifetime prediction models are categorized into mechanistic models, data-driven models and hybrid models. However, data-driven prediction methods often overlook degradation mechanisms, while mechanistic models lack real-time data integration. Additionally, the incomplete exploration of mathematical degradation mechanisms poses challenges for mechanistic models. The frequent start-stop cycles result in significant voltage recovery phenomena, complicating the accuracy of remaining lifetime predictions. This paper presents accelerated stress tests conducted on a 15kW fuel cell stack under idle, rated, and dynamic load conditions, discussing performance recovery following shutdown periods under various operating conditions. The experimental results indicate significant voltage recovery at all current conditions, with the most pronounced recovery occurring at high current conditions. Voltage recovery increases with longer downtime durations. Additionally, a data-driven model utilizing complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and long-short term memory network (LSTM) was proposed. The model first decomposed raw voltage data into modal sequences with distinct characteristic time scales, which were then input into the LSTM for voltage prediction. The prediction results demonstrate that the CEEMDAN-LSTM hybrid model reduces RMSE by 30.41% and 18.21% compared to the LSTM and GRU models under idle conditions, by 28.37% and 16.87% under rated conditions, and by 17.02% and 16.14% under variable load conditions, respectively.

Keywords Proton exchange membrane fuel cell, reversible degradation, degradation prediction, CEEMDAN-LSTM