基于多物理量融合图自编码器网络的质子交换膜燃料电池非均匀反应预测

Abstract To meet the demands of high power output , proton exchange membrane fuel cells (PEMFCs) with large area have become a significant focus of research. However, non-uniform reactions in fuel cells are unavoidable in practice, leading to performance degradation and reduced stack lifespan. Understanding the distribution of physical quantity changes within the fuel cell and predicting its future internal states are crucial for control and maintenance of fuel cells. This paper proposes a Multi-Physics quantity fusion Graph Auto-Encoder network (MP-GAE), which is a transient prediction model for the performance and multi-physical field distribution in fuel cell by focusing on three aspects: reaction time, spatial location , and the coupling relationships of multiple physical fields. Based on graph attention mechanisms and temporal networks, a Partitioned Temporal Graph Attention Network (PT-GAT) is established to extract spatiotemporal regularities. Based on the Auto-Encoder structure and the interrelationships among the five physical quantities, these prediction models are integrated into MP-GAE to enhance the model's prediction performance. Experimental results show that MP-GAE can accurately predict changes in physical fields and performs well under complex conditions such as variations in load current density, gas pressure, inlet relative humidity , stoichiometric ratio and temperature. The proposed model effectively predicts the non-uniform variation processes of five physical quantities within the reaction area of fuel cells, providing information and assistance for the control and management of large-area fuel cells.