The accuracy of the voltage model for hydrogen fuel cells is crucial for system performance evaluation and control strategy design. Semi-empirical models are widely adopted due to their clear physical interpretability and concise structure. However, the precise identification of their key parameters (such as activation coefficients, ohmic resistance, concentration polarization parameters, etc.) faces significant challenges including strong nonlinearity, multivariable coupling, and experimental noise interference. Traditional identification methods (e.g., least squares) are prone to converging to local optima or failing to converge under complex operating conditions.To address the above issues, this paper proposes a parameter identification method based on a Backpropagation (BP) neural network. Leveraging the powerful nonlinear regression and prediction capabilities of the BP network, we construct a mapping relationship where operational conditions (temperature, pressure, voltage, current, etc.) serve as inputs and the model parameters to be identified serve as outputs. By collecting multiple sets of steady-state experimental data to train the network and dynamically optimizing the weights using the gradient descent algorithm, the network autonomously learns the complex association patterns between the parameters and operating conditions. Data normalization and cross-validation strategies are incorporated during training to enhance the model's generalization capability and robustness.Simulation results demonstrate that this method effectively overcomes the limitations of traditional identification: Across a wide operating range, the prediction error of the voltage model using the identified parameters is below 1%, significantly outperforming conventional optimization algorithms. Furthermore, the BP network avoids complex mathematical derivations and offers high computational efficiency. It thus provides a reliable solution for high-precision parameter identification of hydrogen fuel cell models, holding substantial practical value for enhancing system modeling and real-time control performance.
 

Parameter Identification；,PEMFC；,BP Neural Network