An Efficient Privacy-Preserving Asynchronous Federated Approach for Intelligent Decision Making in Equipment Maintenance
Kangkang Dou, Fan He, Fanshu Shang, Xingyi Li, Yuhao Dong, Jiayu Liu · 2024
Federated learning enables collaborative training of a high-quality global model across multiple participants without sharing raw local data. Its advantages, such as decentralization, data isolation, and high computational performance, have made it a popular research direction in various fields. However, federated learning faces privacy risks during model parameter transmission. Additionally, reducing communication overhead and computational resource consumption while ensuring data privacy protection remains a critical challenge. To address these issues, this paper proposes an efficient privacy-preserving asynchronous federated learning method tailored for equipment maintenance scenarios. Leveraging asynchronous federated learning techniques, this method enables efficient collaborative training among multiple participants. It also employs homomorphic encryption to protect local data information. By utilizing time-series models, we construct a predictive model for equipment spare parts consumption, providing valuable insights for intelligent maintenance decision-making. Experimental results demonstrate that our proposed method has faster convergence. More importantly, the root mean square error decreases by at least 0.004, the mean square error decreases by at least 0.024, and the relative error decreases by at least 0.008 compared to the baseline.