Scholarlink

To maximize a maintenance plan and secure the reliability of sustainable energy systems, the remaining useful life (RUL) of fuel cells is required to be precisely forecast. Based on the NASA Prognostics Center of Excellence data, this paper proposes a Feedforward Neural Network (FFNN) model for estimating the RUL of the fuel cells. It leverages temperature, voltage and current-three key operational parameters-to explore intricate breakdown patterns. To improve the prediction accuracy, the systematic process that contains the feature selection, data preprocessing, and network parameter tuning was employed. The FFNN had smaller mean absolute error (MAE) of 0.029 and root mean squared error (RMSE) of 0.039 than the BPNN, FNN, RNN, and CFNN, which are more traditional models. The potential of the suggested FFNN strategy as a trustworthy technique for real-time fuel cell prognostics is demonstrated in a comparative manner with these existing models.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Remaining Useful Life Prediction of Fuel Cells Using Feature-Optimized Feedforward Neural Network

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Zohirul Alam,
  • Fariya Tabassum

摘要

To maximize a maintenance plan and secure the reliability of sustainable energy systems, the remaining useful life (RUL) of fuel cells is required to be precisely forecast. Based on the NASA Prognostics Center of Excellence data, this paper proposes a Feedforward Neural Network (FFNN) model for estimating the RUL of the fuel cells. It leverages temperature, voltage and current-three key operational parameters-to explore intricate breakdown patterns. To improve the prediction accuracy, the systematic process that contains the feature selection, data preprocessing, and network parameter tuning was employed. The FFNN had smaller mean absolute error (MAE) of 0.029 and root mean squared error (RMSE) of 0.039 than the BPNN, FNN, RNN, and CFNN, which are more traditional models. The potential of the suggested FFNN strategy as a trustworthy technique for real-time fuel cell prognostics is demonstrated in a comparative manner with these existing models.