Research on Persistent Advantage Learning Based Deep Q Network for Computational Offloading in Edge Computing
Cong Wang, Mingmin Zuo, Shuai Liu, Di Li · 2023
Due to the increased data transmission and response latency resulting from offloading computational tasks to the cloud in multi-user device scenarios, we construct a typical three-layer system model. To minimize the offloading delay inherent in this system model, we designed a computation offloading strategy leveraging Deep Reinforcement Learning. We propose an enhanced algorithm named PDPAL-DQN. This approach integrates Persistent Advantage Learning, Double Q- learning, and Prioritized Experience Replay with the Deep Q Network. Simulation results indicate that across various network configurations and with large-scale user devices, PDPAL-DQN enhances system reward by approximately 13%, 6%, and 5% compared to other established strategies.