Degradation and polarization curve prediction of proton exchange membrane fuel cells: an interpretable model perspective

Balancing the performance and durability of Proton Exchange Membrane Fuel Cells (PEMFCs) is a significant challenge. Current data-driven technologies mainly focus on improving the accuracy of performance degradation prediction. However, these methods are often considered as black-box models, where the internal workings of the model are in an unknown state, making it challenging to explain the relationship of each feature with performance/durability. To address this challenge, a novel interpretable model for PEMFC performance degradation and polarization curve is proposed, incorporating feature selection interpretation and black-box model interpretation. The key innovation of this paper lies in transforming the black-box model into a dual interpretable transparent model both before and during the learning process. Additionally, the prediction and interpretation of PEMFC polarization curves are performed. The Boruta algorithm is employed to interpret the significance of each feature's impact on the model, filtering out key feature variables. Based on the Boruta-selected feature variables, the SHapley Additive exPlanations (SHAP) algorithm is used to calculate the marginal contribution of each numerical value to the predicted target. This provides an in-depth explanation of the internal workings of the model, analyzing how each predictive factor influences the model results. Through Boruta feature analysis, it is discovered that the cathode characteristics of PEMFC have a greater impact on performance degradation than anode characteristics. The SHAP algorithm interpretative analysis identifies the optimal flow rates corresponding to the best performance for the cathode hydrogen flow rate and anode oxygen flow rate as 0.090 to 0.13 NLPM and 0.39 to 0.50 NLPM, respectively. This indicates that the algorithmic framework has a certain physical meaning in explaining PEMFC performance degradation.