移动群智感知中基于联邦学习的参与者选择机制

移动群智感知的发展使得一些任务收集的数据量过大，需要在不接收参与者原始数据的情况下评估数据质量并进行参与者选择。针对这一问题，提出一种基于联邦学习的移动群智感知参与者选择机制。考虑参与者智能终端资源水平、所处交互状态构建参与者智能终端资源评价机制，提出基于线性回归和长短期记忆网络的智能终端资源预测模型。通过预训练测试模型，评估参与者提供的数据质量，结合历史任务完成情况建立参与者信誉评价模型，实现对参与者的动态评价选择。仿真实验结果表明，所提的参与者选择机制在任务完成质量、能量消耗、通信轮数及任务完成时间等多方面体现出较好的性能。

Participant selection mechanism based on federated learning in mobile crowd sensing

With the development of mobile crowd sensing, the amount of data collected by some tasks is too large. It is necessary to evaluate the data quality of participants and complete the selection of participants without sharing the original data of participants. This paper proposed a federated learning-based mobile crowd sensing participant selection mechanism. Considering the resource level and interaction state of participants'' intelligent terminals, it constructed the evaluation mechanism of intelligent terminal resources of participants, and proposed a prediction model of intelligent terminal resources based on linear regression and long and short term memory network. It evaluated the quality of the data provided by the participants through the pre-training test model. Combining with the historical task completion, it established a reputation evaluation model of the participants to realize the dynamic evaluation and selection of the participants. The simulation results show that the proposed participant selection mechanism exhibits better performance in various aspects such as task completion quality, energy consumption, number of communication rounds and task completion time.