Degradation Prediction of PEMFC Based on Data-Driven Method With Adaptive Fuzzy Sampling

Durability is one of the concerns in the large-scale application of proton exchange membrane fuel cells (PEMFCs). The objective of this article is to propose a data-driven approach to achieve long-term prediction. The echo state network-cycle reservoir with a jump (ESN-CRJ) is an extended network based on an echo state network (ESN). ESN model is used to extract state information in the reservoir and transmit it to the cycle reservoir with a jump (CRJ) for voltage prediction of the stack. In addition, an adaptive fuzzy sampling (AFS) method is adopted to sample the training data in this article. The degradation phenomenon is realized in the stack voltage drop; the place where the voltage drop is rapid contains more degradation information, which needs to be extracted more by the prediction model. Experimental results show that the ESN-CRJ with AFS can be an improvement of 22.02% in long-term prediction under the static current. Under the quasi-dynamic current, the long-term prediction accuracy can be an improvement of 25.06%. Consequently, the proposed approach can achieve good performance in the remaining useful life (RUL) prediction.