Research On IP Network Traffic Estimation Via Temporal Prior Information Guided Sparse Representations

Network traffic data is one of the most critical parameters for network operation as well as network management and control.The main reason is that network traffic data contains a large amount of information that has profound guiding significance for network traffic engineering.Network operation,management,planning,and optimization rely on complete and accurate traffic data.However,in actual network traffic engineering,it is costly to directly measure all network traffic data,and actual operation is not feasible,especially for large-scale backbones.Web environment.The general solution is to sample part of the flow data,and then use the flow data estimation algorithm to further obtain the complete flow data.Therefore,how to efficiently use these sampled data to estimate the complete and as accurate traffic data as possible,so as to efficiently control various network traffic projects,has become a hot research direction.In response to the above problems,domestic and foreign researchers have proposed many advanced network traffic estimation methods at the time.These traffic estimation methods have a wide variety of methods and have shown excellent performance on their respective data sets.However,some of these flow estimation methods directly use rough interpolation methods,and do not fully consider the inherent timing prior information of the flow matrix,which leads to the failure of these methods to achieve satisfactory results in practical applications.In addition,there may be various unknown noises in the sampled flow data,which will also cause the proposed method to be unable to accurately estimate the missing flow value.In order to overcome the above-mentioned shortcomings,this paper introduces the sparse learning theory,models the flow matrix estimation problem as a flow estimation model combining temporal prior information and sparse representation,and progressively proposes two flow estimation methods.The main research contents and innovations of this article are as follows:1)Aiming at the incomplete flow data sampled in the flow estimation problem and the defect of not using the priori information of the timing sequence,this thesis introduces a flow estimation(TPig SR)model combining timing prior information and sparse representation,which can not only effectively use the flow data The temporal correlation of the data can also be expressed through sparse representation using the spatial correlation characteristics of the flow data.Furthermore,the TPig SR problem is solved by adopting the alternating optimization solution(ADMM)method.Finally,experiments were conducted on real data sets,and the results showed that the model performed well on the root mean square error(RMSE)indicator.2)Aiming at the unknown complex noise that may exist in the traffic data,this thesis introduces a network traffic estimation model based on TPig SR with noise robust.The model introduces the Laplace distribution to fit the unknown noise in the complex network environment,thereby reducing the noise sensitivity of the model.For the multiple hyperparameters involved in the model,in order to find the optimal parameter values faster and better,a grid search is used to select the optimal model parameters.Finally,the experimental results on the public data set confirm that the proposed model is better than the current widely used flow estimation methods.