Degradation prediction of PEM fuel cell using LSTM based on Gini gamma correlation coefficient and improved sand cat swarm optimization under dynamic operating conditions

Accurate longevity prediction is crucial for the optimal functioning of Proton Exchange Membrane Fuel Cells (PEMFC). This paper proposes the relative voltage loss rate (RVLR) as a novel aging indicator to overcome the limitations of traditional static indicators like voltage and power, which fall short under dynamic operating conditions. An innovative approach using an LSTM neural network model enhanced by Improved Sand Cat Swarm Optimization (ISCSO) and the Gini Gamma Correlation Coefficient method (GG) is presented for predicting PEMFC degradation across variable conditions. This method first conducts a Gini Gamma correlation analysis, then employs the LSTM network to forge a degradation prediction model for the PEMFC, optimizing the number of neurons in the hidden layer, initial weights, and iteration count through ISCSO. Comparative discussions with eight alternative methods and validations through aging experiments on PEMFC under two different conditions underscore the accuracy of this new approach in various application environments. Specifically, the ISCSO-LSTM model achieves R2 values of 0.9713 and 0.9822, MAE values of 0.0026 and 0.0019, and RMSE values of 0.0050 and 0.0032 across the two different datasets, respectively. These results demonstrate the robustness, accuracy, and reliability of the proposed method for accurate degradation prediction.