Boosted deep neural network model for forecasting the electrochemical impedance of a proton exchange membrane fuel cell under varying operating conditions

Accurate predictive models of electrochemical impedance are indispensable for optimizing the performance and efficiency of proton exchange membrane fuel cells (PEMFCs). In this study, the mechanisms governing PEMFC electrochemical impedance under varying operating conditions were comprehensively investigated, and a precise predictive model was developed using a Boosted Deep Neural Network (BDNN). The results indicate that elevated temperatures and pressures effectively enhance proton conductivity and reaction kinetics at low current densities, thereby reducing impedance. Substantial charge transfer resistance and limited proton conduction at low current densities lead to high impedance, whereas moderate current densities facilitate improved reaction kinetics and enhanced charge transfer processes. With 200 base learners, a learning rate of 0.02 and 32 neurons in the hidden layer, the BDNN model achieves an optimal compromise between accuracy and stable convergence, evidenced by mean squared errors (MSEs) of 4.08 × 10−6 (training set) and 5.48 × 10−5 (test set), and mean absolute errors (MAEs) of 9.45 × 10−4 and 2.55 × 10−3, respectively. Prediction errors remain below 5 % in most cases, underscoring robust generalization and strong predictive capability, even in scenarios involving rapid impedance fluctuations. These findings offer a valuable data foundation and reliable modeling framework for in-depth investigations of PEMFC electrochemical impedance mechanisms and for accurate performance prediction.