Research On Intelligent Prediction Method Of Network Traffic Based On Spatiotemporal Characteristic

In the military field,the iteration of combat concepts,the innovation of combat styles and the emergence of combat forces are driving the rapid transformation of warfare from platformcentric warfare to intelligent warfare.In the civilian sector,the rise of technologies such as 5G and the Internet of Things(Io T)has led to the emergence of instantaneous,complex and variable network traffic at key nodes.Therefore,both in the military and civilian fields show characteristics such as highly dynamic changes in network topology,non-linearity and uncertainty of network traffic,which makes the network traffic model unable to fully extract the spatial and temporal characteristics of network traffic.Therefore,in intelligent network systems,accurate and effective network traffic prediction can understand the characteristics and changing trends of network traffic in advance,so as to improve the utilization of network resources and avoid network congestion.1.In the single-node traffic sequence,the traffic in complex network systems is characterized by nonlinear and uncertain characteristics.Aiming at the problems of poor interpretability of deep learning model and inability to effectively extract traffic characteristics when processing insufficient sample data,this thesis proposes a network traffic prediction model(FPKNet)that integrates prior knowledge.This model takes the self-similar characteristics of network traffic as prior knowledge,integrates them into the gating mechanism of LSTM,and then extracts the time characteristics of traffic sequences by combining onedimensional convolutional neural networks and attention mechanisms.Experiments on a WIDE range of open datasets show that the absolute correction coefficient reaches 76.4%,which is at least 10% higher than the traditional statistical model,the prediction ability is better than other existing prediction methods,and the model is interpretable.2.In the multi-node traffic sequence,the high dynamic change of network topology leads to the difficulty in capturing the complex spatio-temporal correlation of traffic.Aiming at the problem that the traditional model cannot fully extract the spatio-temporal characteristics of network traffic,this thesis proposes a network traffic prediction method based on spatiotemporal hybrid graph convolution,which can better extract the spatial and temporal characteristics of the network.The convolution model based on the spatiotemporal mixed graph(STHGCN)is composed of four modules: time slice module,mixed graph module,spatial feature extraction module,and time feature extraction module.Among them,the time-slicing module divides the input time series into three segments to better capture the periodicity of the traffic series.The hybrid graph module combines static graph and dynamic graph,and the adjacency matrix is used as the input of the spatial feature extraction module graph convolution.The time feature extraction module is passed through a gated cycle unit GRU,and finally,the prediction is obtained through the fully connected layer.The model was applied to the Telecom Italia data set.The results show that the STHGCN model can fully extract the temporal and spatial characteristics of network traffic,and has accurate and stable long-term prediction ability,with an absolute correction coefficient of 91.2%,benefiting from the hybrid graph component.