Design And Implementation Of Network Traffic Prediction Subsystem Based On GCN

Network traffic forecasting is an important task in network operation,maintenance and management,and it is the basis for using big data to assist decision-making.With the continuous expansion of the network scale and the increasingly complex network topology,the relevant research on network traffic forecasting is of great significance.At present,the use of spatial-temporal graph neural network models to model,analyze and predict network traffic data with spatial-temporal features has become one of the focuses.The spatial-temporal graph neural network model is a deep neural network model that can perform temporal feature extraction and spatial feature extraction at the same time.By effectively capturing the spatial-temporal dependencies of network traffic,the prediction accuracy can be significantly improved.Due to the complexity and variability of the network topology,the connection relationship between nodes changes dynamically over time.In addition,the heterogeneity of nodes makes each node in the network have different roles and importance.The above characteristics mean that there are many potential spatial dependencies in the network traffic flowing through heterogeneous nodes.And at the same time,a differentiated temporal pattern is formed,that is,the regularity or trend that often occurs in the network traffic time series.When the traditional GCN model is applied to the field of network traffic prediction,it is difficult to effectively model the complex variability of network topology and the heterogeneity of nodes,and it is impossible to deeply mine the complex spatial relationship of network traffic.In addition,network traffic has a distinct periodicity in the time dimension,which is usually manifested in different time granularities(hours,days,weeks)as cycles,and the traffic temporal patterns are highly similar.However,the existing network traffic prediction methods do not consider the simultaneous extraction and effective fusion of network traffic periodic features at different time granularities,resulting in the problem of insufficient extraction of network traffic periodic features.Therefore,in response to the above problems,this paper first designs and implements a GCN-based network traffic prediction algorithm,and then designs and implements a GCN-based network traffic prediction subsystem.The specific results of this research are as follows:(1)This paper proposes a GCN-based network traffic prediction method.First of all,in view of the problem that the traditional GCN model cannot effectively model the complex variability of network topology and the heterogeneity of nodes,a spatial feature extraction method based on multi-representation graph convolutional network is proposed.In this method,a correlation matrix and a self-learning relationship matrix for describing complex spatial relationships are proposed.At the same time,a subgraph partition method based on the node centrality measure and a multi-representation graph convolution based on pathGCN are proposed to jointly optimize the extraction effect of spatial features.Then,aiming at the problem that traffic periodic features at different time granularities are not effectively processed at the same time,a network traffic periodic feature extraction method based on historical correlation embedding and temporal attribute embedding is proposed.Among them,the historical correlation embedding constructs a historical correlation convolution module to mine the temporal correlation between the historical sequence and the current sequence,and proposes a parameter fusion mechanism to effectively fuse the periodic features of network traffic at different time granularities.Temporal attribute optimizes the extraction of periodic features by encoding the representation of traffic in different time units to effectively associate temporal patterns with time periods.Finally,a simulation experiment is carried out on the algorithm,and the proposed method has achieved the best prediction effect on the classic network traffic dataset,effectively realizing high-precision network traffic prediction.(2)This paper designs and implements a GCN-based network traffic prediction subsystem.First of all,the detailed requirements analysis of the system is carried out around the two aspects of function and performance,and the requirements of each functional module of the system and various performance standards are clarified.Secondly,starting from the system architecture design and the workflow of each module of the system,the outline design of the system is introduced in detail.Then,according to the actual application scenarios of network traffic prediction tasks,this paper designs and implements a series of functional modules such as traffic data summary,traffic data collection,traffic data preprocessing,traffic prediction model training,and traffic prediction result display.Finally,the entire subsystem is fully functionally and performance tested.The test results show that the GCN-based network traffic prediction subsystem designed and implemented in this paper can efficiently and reliably realize the functions of the above modules,and meet the requirements of the subsystem in terms of function and performance.