Task Prediction Based Computation Offloading Over Multi-UAV MEC Network

Due to the rapid development of the Internet of things and the access of large-scale intelligent terminals,static edge nodes can’t effectively solve the conflict between device with insufficient resources and intensive tasks.As a mobile base station,UAV can quickly communicate and process equipment tasks in areas with insufficient communication facilities or natural disasters such as earthquakes.However,the offloading research based on UAV base station mostly ignores the important value of historical data and the regularity of tasks generated by terminal device.In addition,the existing research rarely considers large-scale task offloading,and does not consider the change of the number of UAVs required due to the change of tasks generated by terminal devices.This paper establishes a multi UAV assisted mobile edge computing(MEC)network to provide fast and efficient services for large-scale terminal devices.Considering the limited energy of terminal device and UAV and the delay constraints of sensitive tasks,this paper proposes a joint task prediction(TP)and differential evolution(DE)optimization framework,called Tp De Ras.Taking the real traffic monitoring scenario,the UAV deployment and resource allocation are optimized in binary offloading mode(Tp De Ras-1)and partial offloading mode(Tp De Ras-2)to reduce the total cost of the system.In order to obtain the future task sequence set of device,this paper first proposes a TP algorithm based on distributed multi-layer long short-term memory(LSTM)neural network,which realizes the task prediction of different terminal device according to the task generation rule of device.Secondly,based on the prediction task set,the problem of minimizing the total system cost is divided into UAVs deployment subproblem and resource allocation subproblem for joint optimization.This paper uses adaptive DE algorithm to optimize the number and position of UAVs in real time,so as to optimize the deployment of UAVs.In the evolution process,the percentage of completed tasks is taken as the constraint condition,the number of UAVs is optimized by adaptive adjustment algorithm and the position of UAV group is optimized by DE algorithm.At the same time,based on the prediction task set and UAVs deployment scheme,this paper can flexibly use Tp De Ras-1 framework or Tp De Ras-2framework to optimize UAV resources allocation,so as to obtain the most satisfactory computing offloading scheme.Experimental results show that the Tp De Ras framework has a great improvement in performance compared with other offloading algorithms.Tp De Ras scheme can be applied to real scenes with dynamic changes.When the of task data volume is low,the efficiency of Tp De Ras-1 scheme is higher.When the task data volume is high,the total processing cost of the Tp De Ras-2 scheme is lower.