Degradation prediction of proton exchange membrane fuel cell using a novel neuron-fuzzy model based on light spectrum optimizer

Proton exchange membrane fuel cell (PEMFC) is regarded as the most promising clean energy to address the fossil energy crisis and environmental pollution. However, it is susceptible to frequently variable load and the impurities of hydrogen, which can directly cause the degradation of performance over time during operations. Degradation prediction has received much attention in recent years, as it can improve the durability and reliability of the PEMFC system. This paper proposes an effective multi-step-ahead prediction for PEMFC degradation under various operational conditions by using variational mode decomposition (VMD), a double recurrent fuzzy neural network (DRFNN), and a light spectrum optimizer (LSO). The integrated method enables precise prediction of degradation trends of PEMFC using historical testing data, which brings together their advantages. To better learn degradation trends, VMD is applied to decompose the input voltage signal into a series of sub-signals with a simpler structure. Then, DRFNN with a feedback loop is developed to train each sub-signal model, which can learn and memorize past information. To further enhance the prediction precision of degradation model, LSO is adopted to automatically update the network's weights. Finally, the prediction performance of the proposed method is experimentally verified under different load conditions. Compared with other degradation methods, the test results reveal that the proposed method can achieve significant improvements in terms of multi-step-ahead prediction accuracy and robustness.