【摘要】 针对如何在保护无人驾驶车辆隐私安全的同时，保证缓存内容的新鲜时效性的问题，提出一种基于异步联邦学习的内容流行度预测缓存方案。该方案在异步联邦学习算法框架下，根据无人驾驶车辆的速度和位置对参与模型训练的车辆进行选择，减小车辆无法及时上传模型的损失；通过筛选通信价值高的车辆，以支持车辆的高速移动特性，改善全局模型的准确性；引入伪评级矩阵动态预测内容的流行度，保证已缓存内容的新鲜度，并通过节省大量的训练时间提升效率。仿真结果表明：所提方案在缓存命中率方面的性能优于其他基线缓存方案。更多还原

【Abstract】 Aiming at the problem that how to ensure the freshness and timeliness of cached dynamic content while protecting the privacy and security of unmanned vehicles, a content popularity prediction caching scheme based on asynchronous federated learning was proposed. Firstly, under the framework of asynchronous federated learning algorithm, the scheme selected vehicles participating in model training according to the speed and location of driverless vehicles to reduce the loss caused by vehicles failing to upload models in time. Secondly, vehicles with high communication value were selected to support the high mobility of vehicles and improve the accuracy of the global model. Finally, the pseudo rating matrix was introduced to dynamically predict the popularity of content, ensure the freshness of cached content, and improve efficiency by saving a lot of training time. Simulation results show that the performance of cache hit ratio of the proposed scheme is better than that of other baseline cache schemes.更多还原