Research On Adaptive Offloading Mechanism Based On Network Perception In Heterogeneous Network

With the rapid development of the Internet of Things(Io T)and Artificial Intelligence(AI),various intelligent applications have emerged.Although the computation offloading technology of Mobile Edge Computing(MEC)can fully utilize MEC resources to assist resource-limited intelligent terminal devices,the current situation of device-intensive deployment,interconnection of new and old networks,and coexistence of new and old network protocols has exacerbated the heterogeneity of edge networks.This makes it difficult for computation offloading to achieve the expected benefits.Therefore,how to provide timely and reliable services to users at the lowest possible cost in a heterogeneous edge network has become an urgent problem to be solved.In response to the above problem,this thesis proposes a networkaware adaptive offloading mechanism and conducts research from the following two aspects:(1)In response to the challenge of network state perception caused by the high diversity of intelligent application services and the significant randomness in user behavior,which leads to large traffic jitter and strong burstiness,this study proposes a traffic prediction model based on Graph Convolution Network(GCN)to perceive network states from a traffic perspective.The model takes three types of historical data representing different traffic characteristics and feeds them into attention modules and spatio-temporal modules with the same structure for representation learning.It dynamically captures the spatio-temporal correlation of traffic to more accurately predict future network traffic。By employing this approach,the proposed model outperforms GNN-LSTM by 2.71%,3.13%,and 6.72% in three different prediction time intervals,thus enhancing network perception.These results lay the foundation for further research on adaptive offloading mechanisms in subsequent sections.(2)This thesis proposes a self-adaptive offloading mechanism for edge networks with heterogeneous and diverse types of intelligent application services,where it is difficult to achieve effective computation offloading due to the significant differences in personalized requirements.The mechanism is empowered by a Software Defined Network(SDN)within the MEC network,which leverages the feature of SDN’s centralized control to reduce network redundancy and alleviate the coupling between heterogeneous networks.Based on network awareness,the proposed mechanism incorporates predicted latency into the definition of computational tasks,ensuring that terminal devices can adaptively adjust offloading decisions according to the current network status,their own status,and personalized requirements to reduce the overall cost of task completion for terminal devices.