Proton Exchange Membrane Fuel Cells Lifetime Estimation Using GCN-GRU: Simulation and Prospective Tramway Applications

Proton Exchange Membrane Fuel Cells (PEMFC) are a high-efficiency, clean energy source with significant potential for intelligent transportation systems, such as tramways. However, accurately predicting the lifespan of these fuel cells remains a significant challenge, critical for ensuring reliable and continuous tramway operation. This paper proposes an innovative hybrid neural network model combining Graph Convolutional Networks (GCN) and Gated Recurrent Units (GRU) for precise Remaining Useful Life (RUL) prediction of PEMFC. The model employs a graph learning layer to capture inter-node relationships from PEMFC data, constructing an asymmetric adjacency matrix that reflects the system’s internal directional dependencies. It then utilizes a mix-hop propagation layer to integrate time-series data, effectively capturing the dynamic behaviors and performance variations of PEMFC. Experimental results show that this model outperforms traditional GRU and CNN-GRU models in both short-term and long-term predictions, providing more accurate RUL estimations. The model utilizes high-fidelity test data from a France laboratory to improve the accuracy of fuel cell lifetime predictions, and the potential applications and experimental validation of the model in future intelligent transportation systems such as trams are discussed. This innovative approach provides a robust framework for predictive maintenance, and provides reliable data support and optimization schemes for practical applications in tramways, enhancing the reliability and efficiency of intelligent transportation systems.