A Computation Offloading Method for Multi-UAVs Assisted MEC Based on Improved Federated DDPG Algorithm
Chunlin Li, Kun Jiang, Guangxuan He, Fan Bing, Youlong Luo
Abstract
Computation offloading in UAV-assisted MEC has been an interesting research issue. However, in the temporary hotspot areas with traffic explosion, effective solutions that support user privacy security and higher quality of service (QoS) are still lacking. Compared with the previous works, we first constructed a distributed federated learning-based computation offloading method for user privacy security to maximize the normalized weighted sum of task response delay and user energy consumption. We formulated a computation offloading problem considering the UAV hover position, resource allocation, computation offloading decision, and user transmission power allocation, which is a challenging time-series mixed integer nonlinear programming problem. Then, we proposed an improved federated deep deterministic policy gradient (IF-DDPG) algorithm to address it effectively, which improved the traditional federated DDPG algorithm in empirical pooling and exploring noise. Finally, experiment results show that the IF-DDPG can significantly decrease the task response delay and user energy consumption.