Task offloading, underpinned by Unmanned Aerial Vehicle (UAV) relays within the realm of edge computing, has evolved as a vital mechanism to circumvent limitations associated with terminal computing capabilities. Contemporary offloading strategies predominantly prioritize either the minimization of energy consumption in individual UAVs or the adjustment of flight trajectories to accommodate a more expansive pool of mobile users. These strategies often overlook the mechanisms through which UAVs could effectively collaborate to optimize system resources. Given the energy constraints inherent to UAVs, adopting an appropriate cooperative offloading mechanism can ameliorate dilemmas tied to excessive loads or energy insufficiency in a single UAV. As a result, data offloading from the UAV swarm to edge servers can be maximized, mitigating resource wastage during inter-UAV task transfers. This paper introduces a Deep Q-Network-based Resource Allocation Policy (DRAP) tailored to manage task offloading upon the arrival of task data at the UAVs. This policy employs a Deep Reinforcement Learning (DRL) network for task data allocation, delegating suitable UAVs to perform offloading actions. The offloading strategy conducts a central assessment of the current state of all UAVs in the swarm and all pending task data, selecting the most suitable UAV for each task offloading operation. This method significantly alleviates load pressure in scenarios where a single UAV is overly burdened or energy-deprived. Given that local task data is not processed but merely transmitted, task offloading can be considered as a discrete action. This perception enables the transformation of the UAV task offloading challenge into a selection problem within a discrete action space, aimed at maximizing the utilization of UAV resources. Extensive simulations demonstrate the proposed solution beats the benchmark algorithms by providing higher task offload volume and lower communication resource consumption.
Deep Reinforcement Learning Based Task Offloading for UAV-Assisted Edge Computing
13.10.2023
736172 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Deep Reinforcement Learning Based Computation Offloading in UAV-Assisted Edge Computing
DOAJ | 2023
|A Reinforcement Learning Based Task Offloading Scheme for Vehicular Edge Computing Network
Springer Verlag | 2019
|