Evolutionary Deep Q Network for Collaborative Edge Caching
Ming Zhao, Zhenfeng Sun, Mohammad Reza Nakhai · 2022 IEEE Globecom Workshops (GC Wkshps) · 2022
In cache-assisted wireless edge networking, the problem of collaborative caching among various scenarios with different content popularity and dynamic user demands remains to be solved. In this paper, we address this problem and introduce a collaborative edge caching scheme for a mobile edge computing (MEC) system with various caching scenarios, where the content popularity and users’ preference are time-varying and unknown in advance. The updating process of the proposed collaborative edge caching policies is formulated as a multitasking optimization problem and solved by the multifactorial evolutionary solver, based on the multifactorial evolutionary algorithm (MFEA). We develop the evolutionary deep Q network (eDQN) algorithm to collaboratively optimize the caching policies by simultaneously solving multiple optimization tasks at distributed edge servers. The performance of the proposed algorithm is evaluated based on a real-world dataset from MovieLens and compared against three popular caching policies. The numerical results confirm a superior performance of the proposed eDQN algorithm over these existing policies.