Abstract: Traditional linear analysis method takes the network traffic time series analysis,and does not fully use the exi-sted non-linear feature information．The data analysis performance is limited as result．An improved testing model of network traffic with phase space reconstruction based on recurrence quantitative analysis(RQA)was proposed,and on the basis of phase space reconstruction,the recurrence rate called REC feature was designed as the sequence data support．The average mutual information algorithm and false nearest neighbors algorithm were proposed in calculating the key parameters for phase space reconstruction．The determinism and prediction property of network traffic were tested based on the regular points and lines,and the abnormal flow feature was detected based on REC recurrence feature．Simu-lation result shows that the series has strong stability and self-similar characteristics,and the feature detection performance is stable and precise,and the monitoring precision is improved by 19% with REC feature,and the abnormal traffic detection precision is 99.7% with the improvement of 13.2%．It shows good performance in analysis of network traffic and non-stationary data sequence.

Key words: Network traffic,Recurrence rate,Phase space reconstruction,Recurrence plot