User Mobility Feature Learning And Computation Offloading Optimization In Radio Access Networks

In recent years,more and more applications require low latency and high processing capacity.The limited computing capacity of mobile devices limits their performance in complex mobile applications.Opportunistic computation offloading can reduce computing burden and prolong battery life of mobile devices.However,the computation task load on the edge servers becomes unbalanced due to user mobility.Therefore,it becomes an urgent problem to detect user mobility anomaly and balance the computation offloading traffic based on user mobility.In this thesis,the problem of user mobility anomaly detection and collaborative computing offloading optimization is investigated in the scenario of radio access networks.The main research work and contribution includes:(1)in the aspect of user mobility anomaly detection,considering the impact of user mobility privacy and communication overhead on learning data,two federated learning-based algorithms are proposed named Graph attention network based User Mobility Anomaly Representation and Detection(GUMARD).Simulation results show the efficiency of the proposed algorithms.(2)In order to solve the load balancing problem of computation offloading,user mobility is predicted based on Long Short-Term Memory network.Based on multi-agent cooperative learning,two computation offloading optimization algorithms based on user mobility prediction are proposed,including Base Station Collaborative Computation Offloading Optimization Algorithm(BSC-COOA)and Base Station Non-collaborative Computation Offloading Optimization Algorithm(BSN-COOA).The proposed algorithms are evaluated by python,and the performances of two algorithms are verified compared with baseline algorithms.