Degradation Prediction of Proton Exchange Membrane Fuel Cell Under Dynamic Loading Conditions Using Algorithm‐Optimized Long Short‐Term Memory Models

Long short‐term memory (LSTM) and bidirectional long short‐term memory (Bi‐LSTM) neural networks are utilized to predict the performance degradation of proton exchange membrane fuel cells using a dataset of 550 h obtained from the Federation for Fuel Cell Research (FCLAB). The model performance is optimized using Bayesian optimization long short‐term memory (BO‐LSTM), ant colony optimization long short‐term memory (ACO‐LSTM), Bayesian optimization bidirectional long short‐term memory (BO‐Bi‐LSTM), and ant colony optimization bidirectional long short‐term memory (ACO‐Bi‐LSTM). For each above model, how different parameters such as the number of hidden nodes, dropout rate, learning rate, and number of hidden layers affect the prediction results are carefully examined. The results show that the mean absolute percent error (MAPE) of BO‐LSTM, ACO‐LSTM, BO‐Bi‐LSTM, and ACO‐Bi‐LSTM models with a training rate of 80% are approximately 0.000230, 0.000394, 0.000218, and 0.000181, respectively. Note that the obtained MAPE values are improved significantly as compared with those (MAPE = 0.0518–0.0097) of previous reports, leading to the potential application of proposed models for similar fuel cell systems.