Privacy-preserving predictive maintenance method for cross-border unmanned logistics system integrating federated learning and blockchain

Predictive maintenance of cross-border unmanned logistics systems (CBULS) faces multiple challenges such as data privacy protection, system performance optimization, and collaborative efficiency. To solve these problems, this paper proposes a predictive maintenance method that integrates privacy-preserving federated learning and dynamic consensus blockchain. In the federated learning part, the improved FedProx algorithm is used to deal with non-independent and identically distributed (non-IID) data and device heterogeneity, and multi-layer privacy protection mechanisms such as zero-knowledge proof, fully homomorphic encryption, and local differential privacy are introduced to enhance data security. In the blockchain part, a hybrid consensus mechanism combining delegated proof of stake (DPoS) and practical Byzantine fault tolerance (PBFT) is designed to achieve secure distributed collaboration in high-throughput and low-latency scenarios. In addition, the hierarchical structure and sharding technology are used to optimize system performance and improve algorithm scalability and computational efficiency. Test results show that this method is superior to existing methods in terms of model prediction accuracy, communication efficiency, system throughput, and privacy protection strength, providing an efficient, secure, and scalable solution for predictive maintenance of CBULS.