Research on Algorithms of Computing Offloading and Resource Allocation Based on DQN
Guoqing Chen, Qiong Cai, Xingbao Fu, Yuanzhe Lai · Journal of Physics Conference Series · 2021
Abstract In the MEC unloading system with the weighted and minimized delay energy consumption as the optimization goal. Firstly, build a network model consisting of mobile devices, MEC servers and cloud centers. Secondly, considering the composition of the task, the amount of input data, the energy consumption threshold, the delay threshold and other factors, the task is divided into a non-unloadable part and an unloadable part, and then the unloadable part is divided into several subtasks. According to the AHP level analysis, this method Assign weights to subtasks to determine the allocation of computing resources. Finally, combined with the deep reinforcement learning method to determine the offloading decision, consider the frequency of the same task, and store the high-frequency tasks in the cloud center. In the simulation experiments of different system parameters, the results show that the combination of the model and the AHP-DQN method can better reduce the total system cost.