Research Article
Collaborative Mobile Edge Caching Strategy Based on Deep Reinforcement Learning
@INPROCEEDINGS{10.1007/978-3-030-63941-9_1, author={Jianji Ren and Tingting Hou and Shuai Zheng}, title={Collaborative Mobile Edge Caching Strategy Based on Deep Reinforcement Learning}, proceedings={6GN for Future Wireless Networks. Third EAI International Conference, 6GN 2020, Tianjin, China, August 15-16, 2020, Proceedings}, proceedings_a={6GN}, year={2021}, month={1}, keywords={Mobile edge caching Deep reinforcement learning Federated learning}, doi={10.1007/978-3-030-63941-9_1} }- Jianji Ren
Tingting Hou
Shuai Zheng
Year: 2021
Collaborative Mobile Edge Caching Strategy Based on Deep Reinforcement Learning
6GN
Springer
DOI: 10.1007/978-3-030-63941-9_1
Abstract
Recently, with the advent of the 5th generation mobile networks (5G) era, the emergence of mobile edge devices has accelerated. Nevertheless, the generation of massive edge data brought by massive edge devices challenges the connectivity and cache computing capabilities of the internet of things (IoT) devices. Therefore, mobile edge caching, as the key to realize efficient prefetc.h and cache of edge data and improve the performance of data access and storage, has attracted more and more experts and scholars’ attention. However, the complexity and heterogeneity of the devices in the edge cache scenario make it unable to meet the low latency requirements of 5G. In order to make the mobile edge caching more intelligent, based on the widely deployed macro base stations ((\xi )BSs) and micro base stations ((\mu )BSs) in 5G scenarios, the(\xi )BS cooperation space and(\mu )BS cooperation space is conceived in this paper. Besides, deep reinforcement learning (DRL) algorithms with perception and decision-making capabilities are also used to implement collaborative edge caching. DRL agents perform original and high-dimensional observation training on high-dimensional edge cache scenes, which can effectively solve the dimensionality problem. Then, we jointly deployed federated learning (FL) locally to train DRL agents, which not only solved the problem of resource imbalance, but also realized the localization of training data. In addition, we formulate the energy consumption problem in the collaborative cache as an optimization problem. The simulation results show that the solution greatly reduces the cost of caching and improves the user’s online experience.
