汽车工程 ›› 2023, Vol. 45 ›› Issue (3): 393-401.doi: 10.19562/j.chinasae.qcgc.2023.03.006

所属专题: 新能源汽车技术-动力电池&燃料电池2023年

• • 上一篇    下一篇

燃料电池模型多尺度参数双代价函数的全局灵敏度分析

王俊峰,陈吉清,兰凤崇(),刘青山,曾常菁   

  1. 1.华南理工大学机械与汽车工程学院,广州  510640
    2.广东省汽车工程重点实验室,广州  510640
  • 收稿日期:2022-10-21 修回日期:2022-11-13 出版日期:2023-03-25 发布日期:2023-03-22
  • 通讯作者: 兰凤崇 E-mail:fclan@scut.edu.cn
  • 基金资助:
    国家自然科学基金 (52175267)、广州市科技计划项目 (202007020007) 和广东省自然科学基金(2021A15150912) 资助。

Global Sensitivity Analysis of Multi-scale Parameters of the Fuel Cell Model Based on Two Cost Functions

Junfeng Wang,Jiqing Chen,Fengchong Lan(),Qingshan Liu,Changjing Zeng   

  1. 1.School of Mechanical & Automotive Engineering,South China University of Technology,Guangzhou  510640
    2.Guangdong Province Key Laboratory of Vehicle Engineering,Guangzhou  510640
  • Received:2022-10-21 Revised:2022-11-13 Online:2023-03-25 Published:2023-03-22
  • Contact: Fengchong Lan E-mail:fclan@scut.edu.cn

摘要:

质子交换膜燃料电池半经验模型存在较多未知经验参数,且其尺度和取值范围均相差较大,在全局灵敏度分析中局部高灵敏度信息易丢失,从而导致结果存在偏差。为此,构造了互为倒数的双代价函数。在原代价函数基础上计算全域内经验参数与响应误差的相关性,利用Sobol方法均匀采样进行首次全局灵敏度分析,筛选整个取值范围内高灵敏度(即影响收敛速度)参数;进而利用倒数代价函数放大局部区域的误差相关性,对于剩余全域不敏感参数,进行倒数代价函数的再次全局灵敏度分析,筛选出局部取值范围内高灵敏度(即影响收敛精度)参数。从而提高了多尺度多局域高灵敏参数识别能力。分析结果证明了所得高灵敏度参数辨识后模型的响应误差与全参数辨识的结果一致,可节省约60%的计算成本。该方法的适用性和准确性在电堆实验中均得到了有效验证。

关键词: 质子交换膜燃料电池, 半经验模型, 全局灵敏度分析, 代价函数, 参数辨识

Abstract:

The semi-empirical model of the proton exchange membrane fuel cell has several unknown empirical parameters with widely varying scales and ranges of values, which can lead to biased results due to the loss of local high sensitivity information in the global sensitivity analysis. For this reason, double cost functions are constructed that are reciprocal to each other. On the basis of the original cost function, the correlation between the empirical parameters and the response error in the full domain is calculated. A first global sensitivity analysis is performed using the Sobol method for uniform sampling. The parameters with high sensitivity (i.e. affecting the speed of convergence) in the whole range of values are filtered; then the reciprocal cost function is used to amplify the error correlation in the local area, and for the remaining globally insensitive parameters, another global sensitivity analysis of the reciprocal cost function is carried out to filter the parameters with high sensitivity (i.e. affecting the accuracy of convergence) in the local range of values. Thus, the ability of identification of multi-scale and multi-local high-sensitivity parameters is improved. The analysis results show that the response error of the model after high-sensitivity parameter identification is consistent with the results of the full parameter identification, saving about 60% of the computational cost. The applicability and accuracy of the method are verified by fuel cell stack experiments.

Key words: PEMFC, semi-empirical model, global sensitivity analysis, cost function, parameter identification