| With the rapid development of the Internet of Things and 5G wireless networks,the data generated at the edge of the network has grown exponentially.The original method of uploading data to the cloud in a unified manner and centrally processing it by the cloud server has gradually exposed high latency,high cost,Data privacy and security are not guaranteed.The emergence of edge networks has solved the above problems.The edge network consists of a variety of devices,such as portable terminal devices,wireless sensors and other small computing devices.Services are deployed on edge nodes that are closer to users to meet the real-time needs of users and reduce the load on mobile communication networks.In edge network scenarios,due to the increasingly complex service types and component architectures,failures have become the norm in edge network scenarios.The fault prediction technology is an active reliability management.It is of great significance to improve the reliability of the edge network by analyzing the historical state data of the system to predict whether the system will have some kind of fault in the future.Because edge network systems are closer to actual users,the real-time demand for data is also higher.The traditional centralized cloud fault prediction technology cannot be directly applied to edge network systems due to its high computational complexity.At present,some progress has been made in traditional network fault prediction methods,but the fault prediction for edge networks is still in the preliminary stage of research.Traditional network fault prediction mainly uses support vector machine algorithm,clustering algorithm,expert system and other methods,but there are problems such as difficulty in achieving fine-grained prediction,poor learning ability,and inability to accurately perceive the spatiotemporal correlation between operating data.With the development of deep learning algorithms,deep learning has been widely used in traditional communication network fault prediction due to its effective automatic learning mechanism.However,in the specific scenario of edge network fault prediction,deep learning algorithms have the disadvantages of large consumption of computing resources,Defects such as incomplete feature extraction of fault information.In response to the above problems,this paper proposes a lightweight fault prediction algorithm for edge networks,and designs and completes a fault prediction subsystem for edge networks.The specific innovations of this paper are as follows:(1)A Lightweight Fault Prediction Method for Edge Network is proposed,which realizes fault prediction through the cross-layer cooperation of "edge device-edge server".First,on the edge node side,a lightweight temporal feature extraction algorithm based on brain neural fusion LSTM is proposed.The algorithm searches the historical fault state information in the constructed fault information tree,and combines the current state information for feature learning,which improves the learning ability of the time feature of the fault information.Prune,and then achieve the effect of lightweight failure prediction model.Then,on the edge server side,a lightweight extraction method of spatial features based on the twoway residual structure is proposed.This method uses decomposed convolution to reduce the amount of parameters of the network,realizes the lightweight of the model,and integrates the additive information flow and the multiplication information flow.,to improve the learning ability of the model to the spatial characteristics of fault information,thereby effectively improving the effect of fault prediction.By extracting and splicing the temporal and spatial features of the above fault information,fault prediction is achieved.Finally,the simulation experiment of the lightweight fault prediction algorithm for edge network proposed in this paper is carried out.Compared with the traditional communication network fault prediction algorithm,the algorithm proposed in this paper reduces the calculation amount of the algorithm,and the accuracy,recall rate,F1,etc.There has been a certain improvement in the indicators.(2)This paper designs and implements an edge-oriented network fault prediction subsystem with the help of Vue and Spring Boot development frameworks.First of all,analyze the requirements of the system,design the system architecture and the specific work flow of each module.Then,based on the algorithm proposed in this paper,modules such as network overview,node monitoring,model training,fault prediction,and user management are designed and implemented.Finally,the function and performance of the system are tested.The results show that the edgeoriented network fault prediction subsystem implemented in this paper can accurately and efficiently realize network overview,node monitoring,model training,fault prediction,user management and other functions to meet the actual needs of users. |
