面向车联网的联邦学习模型定制框架及算法改进

Customized federated learning model framework and algorithm enhancements for vehicular networks

针对车联网联邦学习服务难以满足用户训练个性化模型的需求,提出一种创新性的车联网联邦学习模型定制化服务框架。该框架采用了一种融合设备贡献度和数据集相似性的联邦学习聚合算法,实现了个性化联邦学习。该算法通过不同权重分配方式和相似性计算,使得不同用户可以根据自己的需求和数据特征,选择合适的模型训练方案。该框架还提出了一种双重抽样验证方法,解决了模型性能和可信度问题;此外,利用智能合约支持数据协作,保障了数据的安全性。实验结果表明,提出算法在大多数实验场景中表现出较高的准确率,该框架可以显著提高车联网服务的个性化水平,同时保证模型的准确性和可靠性。

This paper proposed an innovative customized service framework for the federated learning model of the Internet of Vehicles(IoV),which addressed the difficulty of meeting the needs of users to train personalized models in IoV federated learning services.This framework adopted a federated learning aggregation algorithm that integrated device contribution and dataset similarity,achieving personalized federated learning.This algorithm used different weight allocation methods and similarity calculations to enable different users to choose appropriate model training schemes based on their own needs and data characteristics.The framework also proposed a dual sampling validation method to address model performance and credibility issues,and utilized smart contracts to support data collaboration,ensuring data security.The experimental results show that the proposed algorithm exhibits high accuracy in most experimental scenarios,and this framework can significantly improve the personalized level of IoV services while ensuring the accuracy and reliability of the model.