Abstract: To address the problem of large resource consumption for resource intensive tasks computing in dynamic and time-varying unmanned aerial vehicle (UAV) networks with high transmission delay and low reliable connection, digital twin is leveraged to construct twin networks model, which consists of UAV, ground smart terminal and wireless network environment, in order to simulate the operation state of UAV networks. Furthermore, computing tasks offloading mechanism for ground smart terminal is developed in the constructed twin networks model. The ground smart terminal chooses to offload all the computing tasks to the UAV, or perform computing locally, under the constraint of maximum tolerant computing delay. Then, the problem of computing offloading is modeled as a Markov decision process, and an adaptive resource optimization model is established for jointly optimizing UAV hovering point, computing tasks offloading decision and UAV computing resource allocation, so as to maximize the UAV utility. Moreover, a digital twin model empowered Proximal Policy Optimization (PPO) approach is designed to obtain the optimal solutions. Numerical results illustrate that the proposed approach can effective improve the UAV utility. Meanwhile, it can well adapt to the virtual-real mapping error.

Key words: Digital twin, UAV networks, computing tasks offloading, adaptive resource optimization, PPO