摘要
大规模网络的流量行为体现为一个相当复杂的非线性系统,目前国内外对它的研究还没有成熟的方法.多分辨小波分析能将交织在一起的不同频率成分组成的复杂时间序列分解成频率不相同的子序列.基于小波分解和重构思想,文章将流量过程分解成不同尺度下的小波系数和尺度系数,然后分别重构最高层低频序列和各层的高频序列.从各子频率序列中分离趋势项、周期项和随机项,并对各成分分别分析与建模,最后合成原流量时间序列的流量预测模型.通过CERNET流量实例分析表明该模型的精度高于ARIMA模型.
Traffic behavior in a large-scale network is very perplexing, so far the research on traffic behavior doesnt have a well-rounded method. By multi-resolution analysis, the complex traffic time series can be decomposed into many different frequent components. In the paper, based on the wavelet decomposed and recomposed theory, the backbone network traffic are decomposed wavelet coefficients and scale coefficients of different scales. Then the top layer low frequency and all layers high frequency time series are recomposed. And the trend term, period term and random term can be decomposed from these recomposed frequency series, so every sub-series can be analyzed and modeled separately. Finally, the traffic forecasting model can be built by recomposed the sub-series models. The model is proved through CERNET traffic and its precision is larger than ARIMA model.
出处
《小型微型计算机系统》
CSCD
北大核心
2005年第3期400-403,共4页
Journal of Chinese Computer Systems
基金
国家自然科学基金重点项目(90104031)资助
国家"九七三"项目(2003CB314803)资助
关键词
小波
时间序列
网络流量
分解
wavelet
time series
traffic analysis
decompose
