Research And Implementation Of Computation Offloading And Resource Allocation Algorithm In Mobile Edge Computing Based On Deep Reinforcement Learning

The fifth generation of mobile communications technology(5G)is facing new challenges of explosive data traffic growth and large-scale device connectivity.New 5G network services such as virtual reality,augmented reality,driverless cars and smart grids have higher applications for delays,and these computationally intensive applications also consume a lot of energy,and user equipment can not address these problems.Thus,mobile edge computing(MEC)emerges at the right moment.Mobile edge computing deploys computing and storage resources on the edge of mobile networks to meet the stringent latency requirements of some applications.The user equipment(UE)can offload the whole or part of the computing task to the MEC server for computing through the wireless channel,so as to reduce the delay and energy consumption to obtain a good user experience.The existing traditional optimization algorithms are feasible for solving the problem of MEC computing offloading and resource allocation.However,the slot spacing of the MEC system is very small,and traditional optimization algorithms generally require complex operations and iterations to obtain optimization results,so the traditional optimization algorithm is not very suitable for high real-time MEC systems.The reinforcement learning algorithm is well suited to solve decision problems,such as MEC computation offloading decision problems.Reinforcement learning can create learning experiences and accomplish optimization goals through an attempt-reward feedback mechanism that is different from traditional optimization algorithms.The deep learning algorithm can learn the characteristics of historical data.After the training is completed,it has great efficiency improvement compared with the traditional optimization algorithm.If the data of the traditional algorithm is used for training,the advantages of the two can be combined.Based on reinforcement learning and deep learning,this thesis proposes several MEC computation offloading and resource allocation algorithms for different application scenarios.The main work is as follows:1.Under the single-cell multi-user scenario,the computation offloading problem can be divided into full offloading problem and partial offloading problem whether the task can be split.The system modeling of the two kinds of problems is built with the optimization goal of delay and energy consumption.2.For the full offloading problem,a Q-learning algorithm based on reinforcement learning is proposed.In order to avoid the problem of excessive state space,the DQN algorithm based on deep reinforcement learning is proposed.The simulation results show that these two algorithms can effectively reduce the delay and energy consumption.3.For the partial offloading problem,a deep learning network DNN combining traditional optimization algorithm SQP is proposed,and a comparison experiment between dedicated DNN and general DNN is set up.The simulation results show that both algorithms can reduce the running time up to more than three hundred times with small increase on energy consumption.