A Reversible Degradation Adaptive Avoidance-Driven Echo State Network: Application to Fuel Cell Prognostics

Inadequate stack durability limits the wide commercialization of fuel cell vehicles. A crucial prerequisite in enhancing fuel cell stack durability is the accurate prediction of their remaining useful life. During long-term operation, fuel cell stacks may encounter reversible degradation, which poses challenges to prediction accuracy and undermines remaining useful life estimation. This work attempts to isolate reversible degradation using diagnostic techniques to improve prognostic performance. The proposed prognostic strategy is tested on a dynamic working condition dataset. The results show that the proposed method reduces the symmetric mean absolute percentage error and improves the relative accuracy.