Trust-aware task offloading for cost-effective UAV-based edge computing based on reinforcement learning

Trust-aware task offloading for cost-effective UAV-based edge computing based on reinforcement learning

연구 분야: Networking

학회: Neural Computing and Applications

With the development of drone technology, unmanned aerial vehicle (UAV)-based edge computing has emerged as a promising computing paradigm with broad application prospects. However, the limited endurance of UAVs makes UAV-based edge computing more sensitive to the energy consumption and time delay. In addition, open network environment will lead to various potential adversarial attacks targeting information security, such as location privacy leakage, black hole attacks, etc. To combat these challenges, this paper proposes a trust-aware task offloading framework for cost-effective UAV-based edge computing. Firstly, a trust framework to evaluate the level of information security of the devices is proposed. Reinforcement learning algorithms are employed to optimize the trust values. Secondly, a multi-user edge computing model and cost function are established, transforming the multi-objective optimization problem into a single-objective nonlinear programming problem. Thirdly, a cost optimization model is developed to assess the cost-effectiveness of IoT devices in the UAV edge computing system with the constraints of trust values. Additionally, to address convergence issues and local optima challenges, the Monte Carlo Q-learning algorithm is exercised to ascertain the optimal cost. Finally, the simulation results show that the average cost of this scheme is at least 25.22% lower than other schemes in terms of the number of IoT devices, 6.68% lower than other schemes in terms of the CPU frequency of UAV edge computing server, and 18.21% lower than other schemes in terms of the average size of tasks.