Edge computing dynamic unloading based on deep reinforcement learning

Abstract

This article considers a multiple Node server and multiple user scenario and proposes a joint user association and power allocation optimization scheme to minimize power consumption and queuing delay. Firstly, a network and computational unloading model is established based on the comprehensive consideration of random task arrival and time-varying wireless channels. Then, the optimization goal is to minimize the average long-term service cost. A dynamic computing unloading and resource allocation algorithm based on mixed decision deep reinforcement learning is proposed for this research objective. By calling the Actor part of DDPG and combining the Critc part of DDPG with D3QN, the mixed decision problem in edge node scenarios is solved. The proposed algorithm has better stability and faster convergence compared to baseline algorithms such as DQN. At the same time, under different task arrival rates, the average system service cost of the proposed algorithm is significantly reduced.