Prognostics and Health Management (PHM) is among the most significant and effective technologies to improve the durability of a proton exchange membrane (PEM) fuel cell system. This paper deals with the prediction issues of the degradation trend for PEM fuel cells equipped in a city bus. First, three aging parameters are extracted from a multi-parameter voltage model, and two of them are selected to represent the degradation of electronic and ionic resistance separately for the first time. Then the parameters are initialized by harmony search (HS) algorithm with an improved objective function, and updated by resorting to the particle filtering (PF) algorithm. Subsequently, Bayesian ridge regression (BRR) and Gaussian progress regression (GPR) are utilized to establish the relationship between the operating time and aging parameters. We categorized the input of the regression models into two classes: the total operating time and the cumulative time of four operating conditions. The results indicate that the latter performs better than the former in characterizing the future trend of aging parameters. Moreover, it is observed that BRR is more attractive since its computational time is far less than that of GPR while the mean absolute error (MAE) is no more than 8.5 mV.

预后和健康管理 (PHM) 是提高质子交换膜 (PEM) 燃料电池系统耐久性的最重要和最有效的技术之一。本文研究了城市公交车搭载的 PEM 燃料电池退化趋势的预测问题。首先，从多参数电压模型中提取出三个老化参数，首次选取其中两个分别代表电子电阻和离子电阻的退化。然后通过改进目标函数的和声搜索（HS）算法初始化参数，并通过粒子滤波（PF）算法更新参数。随后，利用贝叶斯岭回归（BRR）和高斯进展回归（GPR）建立操作时间和老化参数之间的关系。我们将回归模型的输入分为两类：总运行时间和四种运行条件的累积时间。结果表明，后者在表征老化参数的未来趋势方面优于前者。此外，观察到 BRR 更具吸引力，因为它的计算时间远小于 GPR，而平均绝对误差 (MAE) 不超过 8.5 mV。