Design And Implementation Of Network Short-term Traffic Forecasting System Based On Combined Model

Network short-term traffic prediction is beneficial to network operators to understand the current network operation,and has important significance in network resource configuration and abnormal detection.The prediction accuracy and running speed of network short-term traffic prediction have an important impact on the effective network management,maintenance and security of operators.Network performance warning is an important part of network monitoring.By predicting network short-term traffic in advance and generating different levels of early warning based on actual traffic and set thresholds,network anomalies can be prevented and prepared in advance to ensure network service quality.The traditional network short-term traffic prediction algorithm has the defects of insufficient precision and long running time.The traditional network performance warning method has the defects of high false positive rate and high false negative rate.In this thesis,the following researches are carried out on these urgent problems:1.The mainstream prediction methods of network short-term traffic are reviewed,and the advantages and disadvantages of each method are analyzed.This thesis introduces the characteristics of network traffic and related theoretical knowledge.Based on the self-similarity of short-term traffic,this thesis introduces the nonlinear Markov theory into the field of short-term traffic forecasting,and proposes a short-term traffic forecast based on Markov model.In view of the non-stationary characteristics of the network short-time traffic time series,the gray system theory and the Markov theory are combined to improve the performance of the model.2.In order to further improve the prediction accuracy,using the deep learning method to fit the nonlinear time series,this thesis proposes the gray Verhulst-Markov-GRU neural network model by combining the gray Verhulst algorithm,Markov theory and GRU neural network.The comparison between LSTM and GRU methods on public datasets and private datasets verifies the validity of the three models presented in this thesis and demonstrates the superiority of the grey Verhulst-Markov-GRU neural network model.3.On the basis of high prediction accuracy of gray Verhulst-Markov-GRU neural network model,an adaptive network performance early warning model is proposed.Taking the predicted result of the combined model as the baseline,combined with the adjusted deviation and intensity factor to generate a 3? three-level warning,it is possible to detect network anomalies in advance and make corresponding decisions.The effectiveness of the warning was verified on the data set provided by ZTE Corporation.A network short-term traffic prediction system was developed,which integrates data pre-processing,short-term network flow prediction,and network performance warning,and passed the acceptance of ZTE Corporation.