HAPB-FL: A Federated Learning and Hierarchical Key Agreement Framework with Adaptive Privacy Budgeting for Privacy Preservation in IoT

Abstract

Aiming at the problems of low efficiency, insufficient security and module fragmentation in the existing privacy protection schemes for the Internet of Things, a new privacy protection model is proposed. By introducing a data sensitivity weighting strategy through adaptive privacy budget allocation, the privacy budget and noise scale are dynamically adjusted. The cross-layer synchronization of root keys through hierarchical logical key tree subgroups and a decentralized architecture adapts to the low computing power requirements of edge devices. These two components simultaneously form a closed-loop synergy with federated learning. The experimental results show that the accuracy of the optimized federated algorithm gradually stabilizes after 100 rounds of communication and reaches 92.5±1.5%. The maximum recognition accuracy, predicted recall rate, and the harmonic mean of the recall rate reach 96.35%, 95.88%, and 96.10% respectively. The evaluation of the fusion model revealed that the overall coincidence degree of the output data of this model with the original data trajectory reached 97.6%. The average training time and the processing time of a single piece of data reached 1.2±0.2 min and 2.3±0.2 ms, respectively. The resource occupation ratio of this model reached a maximum of 51.3%. The above results indicate that the privacy protection model proposed by the research institute meets the requirements of large-scale privacy data processing and performs well in terms of real-time performance and efficiency.