Research On SDN Traffic Situation Assessment And Prediction Technology

With the increasing complexity of the network,traditional network management cannot effectively describe the network traffic status through the unit fusion mechanism,and the global performance of network resources is poor.In order to timely know the traffic situation of the whole network,establishing a network traffic situational awareness system to the network management plays a vital role,which can help the network administrator control the network and maintain its stability to a large extent.The network operation status involves multi-source information data such as throughput,delay,and network utilization,which needs to be comprehensively evaluated by techniques such as data fusion and data mining.Compared with the complexity of the traditional SNMP network distributed measurement system,the SDN controller can achieve centralized monitoring of network traffic data.At the same time,with the outstanding performance of machine learning in the field of data processing,this thesis combined rough set analysis and deep learning to research the evaluation and predict technology of SDN network traffic situation.The main work was divided into the following three points:First,the SDN network traffic situation data collection system was designed.At first,considering the influencing factors of the SDN network forwarding equipment,communication links and network traffic,the initial indicator set of SDN network situation was constructed.Then,by designing the data acquisition function module in the SDN controller,the SDN network platform was built to realize the collection and storage of the situation indicator data.Furthermore,the index data preprocessing,Spearman index correlation analysis and R/S index self-similarity estimation were completed.Second,a new SDN network traffic situation assessment method was proposed.Firstly,the clustering method was used to determine the decision attributes of the situation factor matrix,and the rough set analysis information system was constructed to calculate the situation index reduction and attribute importance.Finally,the unsupervised traffic situation assessment model was generated to quantitatively evaluate the current traffic situation of the SDN network.In general,this algorithm effectively solved the problem that the SDN traffic situation label could not be determined,completed the network situation analysis under the SDN complex network,and opened up a new vision of global state observation for SDN network management.Third,SDN network traffic situation prediction research was importantly carried out.On the basis of the situation assessment,after collecting the network data under the SDN network in a certain period of time,this thesis built time series samples,and applied LSTM model for higher precision prediction research.Meanwhile,the GWO algorithm was used to optimize the LSTM prediction model for enhancing the stability of the model.The experimental results showed that GWO-LSTM has faster iterative convergence speed,better training effect and higher prediction accuracy than LSTM.