Remaining Useful Life Hybrid Prediction for Proton Exchange Membrane Fuel Cell Considering Uncertainty Quantification

YU Xiaoran; XIE Changjun; YANG Yang; ZHU Wenchao; GUO Bingxin

doi:10.13334/j.0258-8013.pcsee.240139

English Version Journal Login

您当前的位置：

首页 >

文章列表页 >

Remaining Useful Life Hybrid Prediction for Proton Exchange Membrane Fuel Cell Considering Uncertainty Quantification

更新时间：2026-10-12

- Remaining Useful Life Hybrid Prediction for Proton Exchange Membrane Fuel Cell Considering Uncertainty Quantification
- Vol. 45, Issue 17, Pages: 6804-6816(2025)
- 作者机构：
  
  1. 武汉理工大学自动化学院,湖北省,武汉市,430070
  2. 现代汽车零部件技术湖北省重点实验室(武汉理工大学),湖北省,武汉市,430070
  3. 材料复合新技术国家重点实验室,湖北省,武汉市,430073
- 作者简介：
- 基金信息：
- DOI：10.13334/j.0258-8013.pcsee.240139
  CLC：
- Published：2025
- 稿件说明：
移动端阅览
YU Xiaoran, XIE Changjun, YANG Yang, et al. Remaining Useful Life Hybrid Prediction for Proton Exchange Membrane Fuel Cell Considering Uncertainty Quantification[J]. 2025, 45(17): 6804-6816.
DOI：

YU Xiaoran, XIE Changjun, YANG Yang, et al. Remaining Useful Life Hybrid Prediction for Proton Exchange Membrane Fuel Cell Considering Uncertainty Quantification[J]. 2025, 45(17): 6804-6816. DOI： 10.13334/j.0258-8013.pcsee.240139.

摘要

老化模型是评估质子交换膜燃料电池(proton exchange membrane fuel cell，PEMFC)健康状态(state of health，SoH)和剩余使用寿命(remaining useful life，RUL)的关键，然而，其因诸多原因导致的不确定性降低了模型精度和可信度。因此，提出一种模型不确定度和SoH同时量化(model uncertainty and SoH simultaneous quantification，MUSQ)算法，用于指导和修正卷积神经网络-长短期记忆(convolutional neural networks-long short-term memory，CNN-LSTM)神经网络混合模型的长期预测，构建全新的RUL混合预测框架。采用动态负载循环耐久性实验数据，将该混合预测方法与扩展卡尔曼滤波算法、自适应扩展卡尔曼滤波算法、MUSQ算法、LSTM神经网络、CNN-LSTM混合模型等进行对比，该方法具有最优的长期预测性能和RUL估计精度。在负载电流为14.85 A的工况下，该方法累计误差分别降低49.64%、61.33%、30.65%、57.00%和52.90%。

Abstract

The aging model is critical for evaluating the state of health (SoH) and remaining useful life (RUL) of proton exchange membrane fuel cells (PEMFCs). However

uncertainty

caused by various reasons

reduces the accuracy and reliability of the model. Therefore

a model uncertainty and SoH simultaneous quantification (MUSQ) algorithm is proposed to guide and correct the long-term prediction of the convolutional neural network-long short-term memory (CNN-LSTM) neural network hybrid model and a new RUL hybrid prediction framework is constructed. Dynamic load cycle durability test data are used

and this hybrid prediction method is compared with the extended Kalman filtering algorithm

adaptive extended Kalman filtering algorithm

MUSQ algorithm

LSTM neural network

and CNN-LSTM hybrid model

etc. This method has the best long-term prediction performance and RUL estimation accuracy. Under the operating condition of a load current of 14.85 A

this method reduces cumulative errors by 49.64%

61.33%

30.65%

57.00%

and 52.90%

respectively.

关键词

Keywords

references

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

A Multi-level Interlocking Control System for Main Steam Temperature of Coal-fired Units Based on Fixed Value Cruise

Estimation of Frequency Control Capability Considering Dynamic Characteristics of Different Renewable Generation Resources

Series-capacitor-transformer Hybrid High Step-down Ratio Buck Converter

SP-type Coreless Flyback Converter With Transient PT Symmetric Characteristics

The impact of green transformational leadership on radical green innovation

Related Author

No data

Related Institution

No data

AI问答

⁰