Improvement Of Local Support Vector Machine And Application In The Network Traffic Forecasting

The wireless network, as a new network access technology, has being widely used attributing to its flexible and convenience. Meanwhile, more and more complicated practice requirements make the network management much more difficult. Therefore how to effectively utilize networks'resources and protect networks'security is currently the hot spot in research. Growing network traffic forecasting techniques have become the key technology to cope with these problems mentioned above. Because the wireless network has more complicated and more unstable than the wired network, the traditional forecasting models are not fitting to wireless local network. Those methods can not display the characteristics of the wireless network, lead to the poor accurate and result in the worse efficiency of the network management. Through wide study and research the short-term forecasting algorithm of networks'traffic based Chaos theory and Support Vector Machine, and then we propose a new short-term wireless network traffic forecasting algorithm (LSVM-SAX-DTW-HQ) which has better forecasted preciseness and stronger adaptability. The main dedications of this thesis are:Firstly, after analyzing and considering the vulnerabilities of our previously proposed LSVM-DTW-K forecasting algorithm, we use Hannan-Quinn Information Criterion which is used for selecting the number of neighbor points to replace the experience method in LSVM-DTW-K. This has effectively improved the forecasting accuracy.Secondly, the SAX method is designed to symbolic the time series when use DTW algorithm to measuring the similarity between a pair of vectors. Which make the time complexity of DTW from O ( n2)lower to O ( n+ N2).Finally, though the experiments, we verified the model and its improvements were efficacious, and it could make forecasting of wireless networks'traffic more accuracy.