Accurate long-step degradation trends prediction and remaining useful life estimation for proton exchange membrane fuel cells

Proton exchange membrane fuel cells (PEMFCs) have gained widespread recognition as a highly promising and environmentally friendly power generation device. Thus, they are extensively applied in the fields of transportation, distributed power generation, and etc. However, the limited lifetime and high cost of long-term operation of PEMFCs pose significant challenges that hinder large-scale commercialization. Recently, the combination of data science and machine learning technologies has received attention from industry and academia. A novel data-driven prognostics method is used to estimate the remaining useful life (RUL) and voltage degradation trends of PEMFCs by learning from historical aging datasets, which can undoubtedly crucial for the prognostics and health management of PEMFCs. To this end, a novel parallel rotating neuron reservoir (pRNR) is proposed to accurately estimate RUL and forecast the voltage degradation trends of PEMFCs, which integrates the advantages of simultaneous computation of multiple reservoirs computing neural networks. Specifically, the effects of different parameters, including prediction horizons and training lengths, on the prediction performance of the model under two aging test datasets are investigated. Finally, compared with other prediction methods, the results demonstrated that the proposed pRNR method has higher prediction accuracy and better long-step prediction capability, achieving a root mean square error of 2.78 × 10⁻⁰² under FC2 with a training length of 700 hours and a prediction horizon of 5000 steps.