摘要
为了在保障车载命名数据网络中信息实时传输的同时,降低用户隐私泄露的风险,设计了面向车载命名数据网络的联邦LSTM(Long Short-Term Memory)主动内容缓存框架,提高了车辆的缓存命中率,提升了用户的驾驶体验和安全性。该框架结合注意力机制构建基于LSTM的内容流行度预测模型,分为部署于车辆端的局部预测模型和部署于路侧单元端的全局预测模型。首先,根据车辆的速度和位置选择车辆,并将全局预测模型参数发送给被选择的车辆。其次,车辆利用本地存储的兴趣包历史请求数据训练局部预测模型,将训练完成的局部预测模型的参数上传给路侧单元。再次,路侧单元依据车辆的移动特性和请求频率,完成基于路侧单元端的全局内容流行度预测模型的更新。然后,车辆利用训练完成的全局预测模型进行内容流行度预测,并根据预测结果进行内容缓存,同时将预测结果上传到路侧单元。最后,路侧单元根据自身和相邻路侧单元的预测结果对内容流行度进行排序,并选择其缓存内容。仿真结果表明:所提框架在城市场景和高速场景中的缓存命中率均优于其他基线方案。
In order to ensure the real-time information transmission in the Vehicular Named Data Networking and reduce the risk of user privacy disclosure,a Federated Long Short-Term Memory(LSTM)Active Content Caching framework is proposed in this paper.This framework is designed to improve the cache hit rate of vehicles and enhance the driving experience and safety of users.This framework combines the attention mechanism to construct an LSTM-based content popularity prediction model,which is divided into a local prediction model deployed at the vehicle side and a global prediction model deployed at the roadside unit side.Firstly,the vehicles are selected based on the speed and location of the vehicles,and subsequently,the parameters of the global prediction model are sent to the selected vehicles.The vehicles then utilize locally stored historical request data for interest packets to train local prediction models,and the parameters of the trained local prediction models are uploaded to the roadside unit.The roadside unit completes the updating of its global content popularity prediction model based on the vehicle's movement characteristics and request frequency.The vehicles utilize the trained global prediction model to predict the content popularity,and cache contents according to the predicted results.The predicted results are then uploaded to the roadside unit.The roadside unit utilizes the prediction results of itself and neighboring roadside units to sort the content popularity and select its cache contents.The simulation results demonstrate that the proposed framework outperforms other baseline schemes in terms of cache hit rate in both urban and high-speed scenarios.
作者
樊娜
李佳龙
高宇昕
张俊辉
王超
FAN Na;LI Jialong;GAO Yuxin;ZHANG Junhui;WANG Chao(School of Information Engineering,Chang'an University,Xi'an 710018,China)
出处
《微电子学与计算机》
2025年第6期86-96,共11页
Microelectronics & Computer
基金
国家重点研发计划(2021YFB2501204)。
关键词
车载命名数据网络
内容缓存
流行度预测
联邦学习
长短期记忆网络
vehicular named data networking
content caching
popularity prediction
federated learning
long shortterm memory
