Performance prediction and optimization method of DIR-SOFC based on GA-optimized BP neural network: A case study of multi-component fuel

The direct internally reformed solid oxide fuel cell (DIR-SOFC) has the advantages of wide fuel adaptability and high power generation efficiency. Rapid performance prediction and optimization methods play a very important role in reducing performance improvement of SOFC. In this paper, a DIR-SOFC performance prediction and optimization method based on GA-optimized BP neural network was proposed. Using multi-component fuel as a case, 2060 analysis samples were established by 3D numerical simulation, and the current density and temperature of the DIR-SOFC under different fuel components were predicted and optimized by the proposed method. The results show that this method has the advantages of strong generalization ability, high prediction accuracy and fast calculation speed. Aiming for higher current density and lower maximum temperature gradient, the method is applied to achieve optimization combination of fuel components (H₂O, NH₃, H₂, CO, CH₄). At an operating voltage of 0.7 V, the optimal fuel ratio is determined as 0.6% H₂O, 25.6% H₂, 29% CO, 29.4% CH₄ and 15.4% NH₃. The current density is 3336 A·m⁻² and the maximum temperature gradient is 169618 K·m⁻¹. In addition, the weight analysis method was used to study the influence degree of fuel composition on power generation performance. It is found that increasing the volume fraction of H₂O and NH₃ in the fuel reduces the power generation performance, while increasing the volume fraction of H₂, CO and CH₄ in the fuel improves the power generation performance. Increasing the volume fraction of H₂O decreases the maximum temperature gradient while other gases increase it. These conclusions are consistent with the results obtained by the prediction method, which proves the consistency of the proposed method with the physical mechanism. This study has guiding significance for optimizing the operating conditions of DIR-SOFC.