Research Of Traffic Situational Awareness And Abnormal Traffic Prediction In Software Defined Network Based On Neural Network

As a new generation of network architecture,SDN has the advantages of control forwarding separation,network application programmability,and centralized control functions.Because SDN networks are more open,they also face more and newer SDN-specific attacks,such as control plane denial of service attacks,link flood attacks,flow rule flood attacks,password guessing attacks,remote application attacks,and so on.Therefore,traffic awareness and anomaly prediction of SDN network traffic have important application value.1)In the SDN network,the complexity of user requirements and the dynamics of the operating state make the operation logic of the network more complex,and the difficulty of network management is greatly improved.In landscape studies of SDN networks,most researchers conduct research on cybersecurity situational awareness and detection.Due to the unity of the network simulation platform and operating environment,there are fewer research contents on traffic situational awareness during normal operation of the network,and the traffic situation evaluation index system needs to be further improved.At the same time,fewer researchers analyze the statistical characteristics of traffic and explain it.①Aiming at the above problems,by abstracting the functions in each level of the SDN network architecture and combining the design concept of Endsley’s situational awareness model,this paper proposes a traffic situational awareness system framework based on SDN network.Among them,the posture elements of the traffic situation indicator system are extracted from the traffic status at the data level and the state at the control level,and a total of 40 indicators affecting the network traffic situation are selected to construct the indicator set.In this study,the correlation between the traffic situation indicators in the index concentration is analyzed by using the Speedman coefficient,and the reasonableness of the selected index is verified.At the same time,this study studies the self-similarity of SDN network traffic based on this index system.②On the other hand,many situation factors in the network jointly affect the situation of network traffic.Aiming at the dynamics and complexity of network traffic situation prediction,this study proposes a traffic situation prediction model based on PCA-BP neural network.After experimental simulation,the model with the smallest error was selected.Comparative experiments show that compared with the traffic index with full characteristics,the accuracy rate of the index system proposed in this paper is the highest 97.1%,which is 29.4%higher than the average result of the control group,and the effective traffic situation prediction is realized.2)In the SDN network,the centralized nature of the control logic makes the SDN control level extremely vulnerable to network attacks,which affects the operating state of the entire network.The impact of various attacks is that network traffic changes abnormally,network security and equipment are threatened,and the assessment results affecting network performance are necessary,so it is necessary to predict and monitor abnormal network traffic.①In view of the problem that the traditional self-similarity analysis method cannot directly perceive the abnormal traffic in the network,this topic proposes an anomalous traffic detection method that does not distinguish between types--abnormal traffic perception based on the variance rate of the Hurst index.The study achieves preliminary predictions of anomalous traffic in SDN networks.②In order to further distinguish the types of abnormal traffic,this project explores the spatial and temporal characteristics of SDN network flow,and designs an abnormal prediction model of SDN network traffic based on spatiotemporal feature extraction.The model extracts the comprehensive spatiotemporal characteristics of the network stream through the neural network learning model.Among them,the convolutional neural network structure extracts the spatial characteristics of the bytes within the packets of the SDN network anomalous flow,and the recurrent neural network structure extracts the time series characteristics between the packets of the SDN network anomalous flow.Aiming at the abnormal traffic of the network under the DoS attack,a CNN-LSTM abnormal traffic prediction model is constructed by simulation experiments.The results show that the accuracy rate of the model proposed by this institute reaches 84%.Compared with the models of single-time features and spatial features,the proposed models are improved by 3%and 5%,respectively,which effectively realizes the abnormal traffic prediction of SDN networks.