汽车工程 ›› 2025, Vol. 47 ›› Issue (11): 2150-2158.doi: 10.19562/j.chinasae.qcgc.2025.11.009

• • 上一篇    

基于动态特征参数和改进GOA-BP神经网络的车用PEMFC退化趋势预测

薛发俊1,2,3,杨继斌1,2,3(),邓鹏毅1,2,3,武小花1,2,3,陈丽1,2,3,王文龙1,2,3,胡怀祥1,2,3   

  1. 1.西华大学,汽车测控与安全四川省重点实验室,成都 610039
    2.西华大学,四川省新能源汽车智能控制与仿真测试技术工程研究中心,成都 610039
    3.宜宾西华大学研究院,宜宾 644000
  • 收稿日期:2025-03-31 修回日期:2025-05-22 出版日期:2025-11-25 发布日期:2025-11-28
  • 通讯作者: 杨继斌 E-mail:yangjibin08@163.com
  • 基金资助:
    国家自然科学基金(52407254)

Degradation Trend Prediction of Vehicular PEMFC Based on Dynamic Feature Parameters and Improved GOA-BP Neural Network

Fajun Xue1,2,3,Jibin Yang1,2,3(),Pengyi Deng1,2,3,Xiaohua Wu1,2,3,Li Chen1,2,3,Wenlong Wang1,2,3,Huaixiang Hu1,2,3   

  1. 1.Xihua University,Vehicle Measurement Control and Safety Key Laboratory of Sichuan Province,Chengdu 610039
    2.Xihua University,Provincial Engineering Research Center for New Energy Vehicle Intelligent Control and Simulation Test Technology of Sichuan,Chengdu 610039
    3.Yibin Institute in Xihua University,Yibin 644000
  • Received:2025-03-31 Revised:2025-05-22 Online:2025-11-25 Published:2025-11-28
  • Contact: Jibin Yang E-mail:yangjibin08@163.com

摘要:

针对质子交换膜燃料电池(PEMFC)剩余使用寿命(RUL)预测中动态工况表征不足及传统优化算法易陷入局部最优的问题,提出一种融合动态特征参数及改进蚱蜢优化算法(IGOA)与BP神经网络相结合的预测方法。首先,通过季节趋势分解方法提取电压数据的季节性分量,同时量化工况周期内的功率波动率,并采用灰色关联度分析筛选关键特征参数。然后,利用IGOA优化BP神经网络的超参数组,构建IGOA-BP神经网络预测模型。最后,基于实车数据和实验室测试数据集验证了模型性能。结果表明,提出的方法具有更高的预测精度,平均绝对百分比误差小于0.06%,能够实现更精确的燃料电池RUL预测。

关键词: 燃料电池, 寿命预测, 动态特征参数, 改进蚱蜢优化算法, BP神经网络

Abstract:

For the problems of insufficient characterization of dynamic operating conditions and the tendency of traditional optimization algorithms to fall into local optima in the prediction of remaining useful life (RUL) for proton exchange membrane fuel cell (PEMFC), in this paper a novel prediction method is proposed that combines dynamic feature parameters with an improved grasshopper optimization algorithm (IGOA) and back propagation neural network. Firstly, the seasonal component of voltage data is extracted through seasonal-trend decomposition method, while the power fluctuation rate during operating cycles is quantified. Then, key feature parameters are selected using grey relational analysis. Subsequently, the IGOA is employed to optimize the hyper parameters of the back propagation neural network and construct the IGOA-BP neural network prediction model. Finally, the model performance is validated by real-world vehicle data and laboratory test datasets. The results demonstrate that the proposed method achieves higher prediction accuracy with the mean absolute percentage error below 0.06%, which enables more accurate RUL prediction for PEMFC.

Key words: fuel cells, life prediction, dynamic feature parameters, improved grasshopper optimization algorithm, BP neural network