OFFLOADING STRATEGY FOR UNMANNED AERIAL VEHICLE-ASSISTED MOBILE EDGE COMPUTING WITH CONTROLLABLE FLIGHT HEIGHT

Aimed at the time cost problem of unmanned aerial vehicle (UAV) assisted mobile edge computing offloading with flight altitude as a variable, a task offloading algorithm based on improved twin delayed deep deterministic policy gradient is proposed. The algorithm used the self-attention mechanism to strengthen the neural network’s attention to key elements and close the local connection of elements. The normalized exponential function was used to solve the problem of undervaluation of state action value. The method of prior experience extraction was adopted to improve the efficiency of network training. Compared with the existing task offloading algorithm, the simulation results show that the time cost of the proposed algorithm is reduced by more than 10%. Under the energy limit of Unmanned aerial vehicle, it is concluded that the task offloading strategy has the minimum delay and maximizes the system stability.