Energy Consumption Optimization of Joint Offloading Computation Based on MD-mUAV-mBS

Cloud-side-device collaborative network is the trend of B5G. Using different types of server-terminal heterogeneous networks to process terminal tasks can effectively improve network service quality, which causes to consume a quantity of energy. In order to meet the KPI requirements of future green communication, a multi-device (MD), multi-unmanned aerial vehicle (mUAV), and multi-base station (mBS) collaborative offloading computing framework is first proposed, which can jointly provide communication and computing services to users. Then, a related energy consumption model for offloading computing network is established. Next, deep deterministic policy gradient (DDPG) algorithm is explored to learn the dynamically changing wireless network information, the location for task execution, and network resources to minimize system energy consumption. Finally, the simulation results showcase the feasibility of the model. Comparisons with the system energy consumption of two benchmark strategies prove that our proposed model is effective and outperforms both benchmark strategies. The influence of different learning rates on the performance of the DDPG algorithm is further analyzed. And it can be concluded that the actor learning rate
$\alpha\pi$
has a great influence on the convergence trend of the energy consumption model for offloading collaborative computing.