A Data-Driven DA-LSTM Approach For Predicting PEMFC Degradation Trends in Dynamic Load Conditions

Ekmekçi N. K., Uyar M.

9th International Hydrogen Technologies Congress, İzmir, Türkiye, 25 - 28 Mayıs 2025, ss.935-938, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Basıldığı Şehir: İzmir
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.935-938
  • Bursa Uludağ Üniversitesi Adresli: Evet

Özet

This study proposes a data-driven dual attention long short-term memory (DA-LSTM) approach for predicting the degradation trends of proton exchange membrane fuel cells (PEMFCs) under dynamic load conditions. Addressing key challenges in fuel cell durability and reliability, the proposed model is compared against traditional long short-term memory (LSTM) architectures to assess its effectiveness in capturing complex degradation patterns. The predictive performance of each model is evaluated using root mean squared error (RMSE), mean absolute error (MAE), and the coefficient of determination (R²). The results demonstrate that DA-LSTM outperforms the conventional LSTM model, offering enhanced feature selection and improved long-term prediction accuracy through its dual-stage attention mechanism. These findings highlight the potential of DA-LSTM for fuel cell prognostics, providing a robust framework for prognostics and health management (PHM) systems and advancing the development of sustainable energy technologies.