Akademik Veri Yönetim Sistemi
International Congress on Advanced Energy Studies-II, New York, Amerika Birleşik Devletleri, 24 - 26 Ağustos 2025, ss.15-27, (Tam Metin Bildiri)
Proton exchange membrane fuel cells (PEMFCs) have emerged as a promising
sustainable energy source due to their high efficiency, zero emissions, and low
operating temperature. However, performance degradation over time remains a
significant challenge to their long-term reliability. Consequently, accurate
estimation of the remaining useful life (RUL) of PEMFCs is essential for
predictive maintenance and safe operation. This study presents a data-driven
methodology employing deep learning models to estimate the RUL of PEMFCs based
on voltage time series data. Two model architectures are evaluated: a
conventional long short-term memory (LSTM) model and an advanced LSTM model
enhanced with an attention mechanism. Both models are trained on time series
data from the IEEE PHM 2014 Data Challenge, which includes over 1,100 hours of
PEMFC aging measurements under steady-state load conditions. Following
preprocessing and filtering, the dataset is downsampled to 1,155 data points
and divided into training, validation, and test subsets. Bayesian optimization
is utilized to fine-tune the model hyperparameters. Experimental results
demonstrate that the LSTM-Attention model surpasses the standard LSTM model,
achieving an RMSE of 0.001457, MAE of 0.0003, and R² of 0.985, compared to the
LSTM’s RMSE of 0.001869, MAE of 0.0004, and R² of 0.979. These findings
underscore the effectiveness of incorporating attention mechanisms to enhance
the accuracy of RUL prediction for PEMFC systems.