Abstract:

In order to quickly and accurately predict the start-up duration of a proton exchange membrane fuel cell (PEMFC) during the cold-start process and the application effect of the start-up method, two real-time prediction models of the stack temperature change trend are proposed. The two prediction models take the stack temperature and the temperature increment as the prediction targets of the BP neural network, named model T and model K, respectively. Meanwhile, four different prediction accuracy evaluation criteria are used to evaluate the accuracy of the prediction results. The prediction model is verified based on the experimental data of three cold-start conditions in the literature. The results show that the average relative error of model K is lower than that of model T under the three cold-start conditions, which are 0.4553, 0.9537, and 1.0844, respectively. Model T lacks training samples in the early prediction stage, and the stack temperature variation trend of the prediction results is zero, so model K has a greater advantage in the early prediction stage. The stack temperature change trend prediction method can provide a reference for the user’s current PEMFC cold-start implementation effect.

Key words: fuel cell, prediction, neural network, cold-start, stack temperature

摘要：

为了快速准确地预测出质子交换膜燃料电池(proton exchange membrane fuel cell，PEMFC)在冷启动过程中的启动时长及启动方法的应用效果，提出了以堆栈温度和温度增量分别作为BP(back propagation)神经网络预测目标的堆栈温度实时预测模型，分别为模型T和模型K，并采用四个不同的预测精度评估标准来评估预测结果的准确性。基于文献中三种冷启动工况实验数据对预测模型进行验证，结果表明，模型K的平均相对误差在三种工况下均低于模型T，分别为0.4553、0.9537和1.0844。模型T在早期预测阶段缺乏训练样本，预测结果的堆栈温度变化趋势为零，因而模型K在早期预测阶段具有更大优势。堆栈温度变化趋势预测方法能够为用户当前的PEMFC冷启动实现效果提供参考。

关键词: 燃料电池, 预测, 神经网络, 冷启动, 堆栈温度