A Weighted Privacy-Preserving Federated Learning for Proton Exchange Membrane Fuel Cell

Proton exchange membrane fuel cells (PEMFCs) are emerging as a promising new energy technology with the potential to significantly transform the field of new energy vehicles. However, due to the limited availability of data at individual institutions—since PEMFCs technology remains in the developmental stage—collaborative learning across multiple institutions becomes essential. Given the high confidentiality of research data and models, institutions are unwilling to share raw data or trained models. The widely accepted approach is to utilize federated learning to balance privacy and data availability. However, existing federated learning schemes cannot meet the requirement of PEMFCs since low accuracy and high centralization. To address the limitations of existing methods, which suffer from centralization and low aggregation accuracy, this paper proposes a decentralized, high-precision, and dynamic privacy-preserving federated learning framework. The proposed approach employs aggregation techniques combined with homomorphic encryption to ensure privacy and decentralization. Unlike traditional federated learning frameworks, our protocol does not rely on any central server or coordination node; all participants are treated equally. The aggregation results are deterministic and not influenced by added random noise, thereby ensuring high model accuracy. Moreover, the system supports dynamic participation, allowing users to join or leave at any time. Experimental results demonstrate that our framework achieves high computational efficiency and robust performance, making it a practical and secure solution for collaborative PEMFC development.