Research On Dynamic Reconfiguration Technology Of Slicing In Ad-Hoc Network

Ad-Hoc is a non-centered,self-organizing,multi-hop wireless network that does not require any communication infrastructure.Because of its simple networking,strong flexibility,and strong survivability,it is used in various harsh environments,such as battlefield communications,earthquake relief,and environmental monitoring.Especially in recent years,"mosaic warfare","cloud operations" and other new cyber warfare concepts proposed in the military application field have put forward higher requirements on the Ad-Hoc network.Under new combat concepts such as mosaic warfare,the network is divided into many small and low-complexity systems,which can be combined and spliced on demand,so as to achieve multi-domain collaborative work and have strong rapid reconstruction capabilities.Network slicing divides the physical network into multiple independent virtual sub-networks,and realizes different network requirements by customizing network functions,which coincides with the concept of on-demand combination in the mosaic warfare.In order to cope with scenarios similar to mosaic warfare,to meet diversified network requirements and flexible and rapid network reconfiguration,the reasearch considers introducing network slicing into the Ad-Hoc network.For this new network architecture,reasonable resource management can improve resource utilization and ensure quality of service.However,since the amount of sliced resources remains unchanged for a period of time after resource allocation is completed,business changes and dynamics of wireless Ad-Hoc network can lead to a mismatch between network requirements and allocated resources.Thus,slice reconfiguration is very important.We applies network slicing to Ad-Hoc network,and studies the dynamic reconfiguration problem of Ad-Hoc network slicing.The main research contents are as follows:First of all,we proposes a reconfiguration mechanism for Ad-Hoc network slices.Slice reconfiguration is divided into inter-slice resource reconfiguration and intra-slice resource reconfiguration.Inter-slice reconfiguration will cause service flow rerouting and changes in the amount of resources of the slices,resulting in large network overhead,and even severely affecting the quality of service,so it needs to be performed carefully in a large time interval.Intra-slice resource reconfiguration reallocates resources for flows without changing the amount of resources owned by the slice,which has a small impact on the slice.The reconfiguration overhead is also small,so we consider reconfiguration of resources within slices on a smaller time scale.Then this research uses the long short time memory neural network algorithm to predict the traffic on the slice,but due to the randomness and suddenness of the business in the network,the point prediction is not very good,and it lacks accuracy norms to measure forecast results.Therefore,the research uses bootstrap method to predict the flow,and obtains the confidence interval of the flow according to the results of multiple predictions.However,the bootstrap algorithm requires running the LSTM many times,and the complexity is relatively high.Therefore,this research models the flow by alpha stable distribution and solves the confidence interval according to its nature.Although the prediction effect is slightly reduced,the complexity is effectively reduced.Next,the research performs resource reconfiguration between slices based on the traffic prediction results,which defines the network benefit as the optimization goal based on satisfaction and reconfiguration cost,and sets up the reconfiguration model with resource constraints and transmission rate requirements as constraints.Because the flow prediction result is an interval,the problem is an uncertainty optimization problem.The research uses robust optimization method to transform the problem,and solves it by the particle swarm algorithm.The simulation shows that the reconfiguration based on the predicted particle swarm algorithm can not only obtain the highest network revenue,but also achieve the maximum throughput and resource utilization.Finally,the research studies resource reconfiguration within the slices.Due to the dynamics of the wireless Ad-Hoc network and the influence between the reconfiguration schemes at different times,the problem is modeled as a Markov decision process.Because of the continuity of the state and action space,the research uses Actor-Critic algorithm,which is one of the reinforcement learning algorithms.The simulation results show that the AC algorithm has the better performance compared with the traditional algorithm.