Aging Prediction of Proton Exchange Membrane Fuel Cells Using Bayesian-optimized CNN-BiLSTM-Attention Networks

Accurate prediction of the lifetime of proton exchange membrane fuel cells (PEMFCs) is essential for enhancing their durability and reducing operating costs. In this paper, a PEMFC aging prediction method based on a Bayesian-optimized convolutional neural network-bidirectional long short-term memory-attention (CNN-BiLSTM-Attention) network is proposed. The method relies only on load current and stack voltage time-series data, eliminating the need for multi-sensor inputs. First, variational mode decomposition reconstructs the original voltage signals by eliminating measurement noise interference. Next, the CNN automatically extracts spatial features, while the BiLSTM learns bidirectional aging trends. Finally, the Attention mechanism prioritizes degradation-related features, and Bayesian optimization refines the hyper-parameters. Experiments are conducted on the PEMFC dataset under two operating conditions, resulting in root mean square error (RMSE) values of 0.0028V and 0.0015V for the voltage prediction, respectively. Performance comparisons with the three other network models indicate that the proposed method reduces the RMSE by 0.0003V to 0.0378V, demonstrating superior accuracy and robustness in real fuel cell health management.