Research On User Task Offloading And Resource Allocation In Edge Intelligence

With the continuous development of mobile devices,new applications continue to emerge,such as augmented reality(AR),gesture recognition,real-time games,etc.These new applications not only require a large amount of computing resources,but are also particularly sensitive to latency.However,due to the size limitation of the user equipment,it is also limited in terms of energy and computing power.Mobile edge computation(MEC)is widely regarded as the key technology to realize the next generation Internet.MEC breaks traditional cloud computing methods and continuously improves user service computing quality(QoS).Compared with mobile cloud computing,MEC reduces the delay that may be caused by data center network congestion.By deploying servers to the edge of the network(such as access points,small base stations,etc.),cloud computing capabilities and IT services are brought closer to the user end.One technology not only meets the requirements of low latency for user equipment,but also reduces energy consumption and prolongs the service life of the equipment.Edge servers provide storage and computing functions at the edge of the network.Edge devices can optionally offload tasks to the edge server for execution.Because the location of the edge server is not clearly defined in the MEC system,and the server's own resources are limited,how to reasonably deploy the location of the edge server and the reasonable task offload and reasonable resource allocation of tasks on the server has become an important issue.At the same time,in the MEC system,the user's location changes in real time,and solving the computational offloading problem in a real-time changing environment is also a challenge for researchers.The research work in this paper is summarized as follows:1)Establish a multi-user and multi-MEC server scenario,and propose a method for server location deployment,task offloading,and resource allocation.A mathematical model is first established in this paper with the goal of minimizing the total energy consumption of all tasks.Secondly,a two-layer optimization scheme is proposed.Considering the real-time changing user environment,the upper layer uses the h-SOM network to optimize the location of the server in real time;because a large number of users bring a huge amount of calculation to the MEC server,we use Improved differential evolution algorithm(IDE)to reduce computational complexity to optimize task offload and resource allocation.Finally,compared with traditional algorithms,simulation results show that the algorithm proposed in this paper can obtain lower energy consumption under different numbers of users,different numbers of servers,and different delays.2)A multi-user and multi-MEC server scenario is also established,and a neural network is proposed to solve the transmit power matrix,task offload matrix,and resource allocation matrix.In this paper,a mathematical model with minimum energy consumption as the objective function is first established,and then an improved differential evolution algorithm is used to collect samples,which is then put into a neural network for training.After training the network,the user's transmit power can be directly output according to the network input.The unloading location and computing resources are reduced for real-time optimization problems.Finally,compared with traditional algorithms,simulation experiments show that the prediction results of the neural network can minimize the energy consumption of the system while making full use of limited resources.