Abstract
With the rapid development of Internet of Things (IoT) technology, the demand for distributed data sharing has surged. Privacy breaches, a single point of failure, and high communication overhead have become core challenges. Existing privacy protection based on single federated learning lacks reliable collaboration mechanisms and fault tolerance capabilities. The traditional PBFT algorithm also faces communication complexity and is incompatible with high concurrency scenarios. To address the limitations of these single technologies, a multi-layer data sharing framework is proposed. It integrates federated learning and blockchain technology is proposed to deal with the security and efficiency issues of privacy data transmission in the IoT environment. The framework uses differential privacy to ensure data privacy and solves server single-point failures using the Proof of Quality (PoQ) consensus algorithm. A reputation mechanism supervises node behavior. Meanwhile, an improved PBFT consensus algorithm based on dynamic region partitioning is designed to handle fast data transmission and improve consensus efficiency in the IoT. By integrating these two solutions, an overall architecture is constructed to achieve efficient IoT data transmission. Experimental verification showed that the differential privacy mechanism only reduced data transmission accuracy by 3.5% after introducing Gaussian noise. In terms of fault tolerance, the PoQ algorithm had a fault rate of 14.3% for fault-tolerant nodes. Dynamic partitioning PBFT optimization, reduced communication frequency by 44% compared to single-layer PBFT, and transaction latency by 63%. These results indicate that the designed solution has significant advantages in high-concurrency IoT scenarios and can effectively meet real-time security demands.