CIESC Journal ›› 2024, Vol. 75 ›› Issue (3): 945-955.DOI: 10.11949/0438-1157.20231247

• Process system engineering • Previous Articles     Next Articles

PEMFC fault diagnosis based on improved TSO optimized Xception

Lingxian ZHANG1(), Bin LIU1,2(), Lin DENG1, Yuhang REN1   

  1. 1.Beijing Sifang Automation Company Limited, Beijing 100085, China
    2.School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
  • Received:2023-11-29 Revised:2024-01-27 Online:2024-05-11 Published:2024-03-25
  • Contact: Bin LIU

基于改进TSO优化Xception的PEMFC故障诊断

张领先1(), 刘斌1,2(), 邓琳1, 任宇航1   

  1. 1.北京四方继保自动化股份有限公司,北京 100085
    2.天津大学电气自动化与信息工程学院,天津 300072
  • 通讯作者: 刘斌
  • 作者简介:张领先(1995—),男,硕士,zhanglingxian@sf-auto.com
  • 基金资助:
    国家重点研发计划项目(2021YFB2400700)

Abstract:

This paper proposes a fault diagnosis method for proton exchange membrane fuel cells (PEMFC) based on Xception network optimized by an improved transient search optimization (TSO) algorithm. First, the fault data are normalized and dimensionally reduced by linear discriminant analysis, which reduces the computational complexity while preserving the main features. Secondly, the TSO algorithm is enhanced by introducing Tent chaotic mapping and reverse learning strategy, which improves its global search ability. The hyperparameters of the Xception neural network are optimized by the TSO algorithm in the training phase. Finally, the fully trained Xception network is used to classify and identify PEMFC faults, and compared with the classic classification model. On the experimental water management fault data and the simulated multi-class fault data, the Xception network achieves the highest classification accuracy, which is 100% and 98.08%, respectively. This indicates that the Xception network has a strong ability to extract data features and the proposed method can serve as a general diagnosis method for PEMFC faults.

Key words: proton exchange membrane fuel cell, fault diagnosis, Tent chaotic mapping, reverse learning, transient search optimization, Xception neural network

摘要:

针对质子交换膜燃料电池(PEMFC)的故障诊断问题,提出了一种利用改进的瞬态搜索优化(TSO)算法优化Xception网络的故障通用诊断方法。首先,对故障数据进行线性判别分析降维和归一化处理,在保留主要特征的前提下降低计算复杂度;其次,引入Tent混沌映射和反向学习策略增强TSO算法的全局搜索能力,在训练阶段对Xception神经网络的超参数进行优化;最后,使用充分训练的Xception网络对PEMFC故障进行分类识别,并与经典的分类模型进行对比。在基于实验测量的水管理故障数据和仿真产生的多类故障数据上,Xception均取得了最高的分类准确率,分别为100%和98.08%,这表明Xception对数据特征的提取能力较强,且所提方法能作为一种PEMFC故障的通用诊断方法。

关键词: 质子交换膜燃料电池, 故障诊断, Tent混沌映射, 反向学习, 瞬态搜索优化, Xception神经网络

CLC Number: