GWO-BP hybrid model optimization for a 110 kW proton exchange membrane fuel cell system considering cathode humidity

Proton exchange membrane fuel cells (PEMFCs) are widely used in heavy commercial vehicles due to high energy density and environmental characteristics. This paper introduces a GWO-BP hybrid model that integrates a grey wolf optimization (GWO) semi-empirical model with a back propagation (BP) neural network error prediction model. This innovative model has the potential to significantly enhance the model dynamic characteristic accuracy of a vehicle-level 110 kW PEMFC system. The GWO semi-empirical PEMFC model is established to accurately represent the physical phenomena of the PEMFC system under steady-state conditions. To improve the dynamic accuracy of the GWO semi-empirical PEMFC model under dynamic load conditions, a BP neural network error prediction model is combined by accounting for the influence of cathode humidity. The results demonstrate that the mean absolute error (MAE) of the output voltage from the GWO-BP hybrid model is 0.78 V, representing a 96.89% improvement in output voltage accuracy compared to the GWO model. Additionally, the MAE of the output power is 0.18 kW, resulting in a 73.91% enhancement in output power accuracy. The GWO-BP hybrid model demonstrates superior performance in system parameter identification and prediction, effectively mitigating the fitting errors of the semi-empirical PEMFC model.