广域子空间异常网络流量预测概率置信度优化

Optimization of Abnormal Network Flow Predictive Confidence Level with Wide Area Subspace

为了提高对网络异常流量监测模型的置信度和准确性,降低虚警概率,防止漏报。需要对异常网络流量预测概率置信度区间进行优化设计。提出一种基于预测流量数据主特征建模与广域子空间重构的异常网络流量预测算法,提高预测概率置信度。把主特征建模投影到广域子空间中,获取网络异常流量数量的对偶子梯度预测结果,其表现为一组尺度,分析网络预测流量的概率置信度的伸缩变化,选用极大似然法对网络流量的预测概率置信度进行特征建模,得到概率置信度区间。提高预测置信度和精度。实验结果表明,该算法能有效提高异常网络流量的预测精度,通过在广域子空间中特征建模,保证了预测的概率置信度,减少预测误差,性能优越。

In order to improve the confidence level and accuracy of network traffic anomaly monitoring model, to reduce the false alarm probability, to prevent the omission. Need to optimize the design of abnormal network traffic prediction probabil?ity confidence interval. An abnormal network traffic flow prediction algorithm to predict the main features of wide area data modeling and subspace is proposed based on reconstruction, design improvement algorithm, the main feature modeling pro?jection to the wide area subspace, to obtain dual sub-gradient number of abnormal network traffic prediction results, its per?formance for a set of scale expansion, analysis the probability of network change flow forecast confidence, use the maximum likelihood method to predict probability network traffic confidence of feature modeling, probability of confidence interval. The prediction confidence and accuracy is improved. The experimental results show that the algorithm can effectively im?prove the prediction accuracy of abnormal network traffic through feature in wide area subspace modeling, ensure to predict the probability of confidence, reduce the forecasting error, it has superior performance.