Forecasting Of Networks' Traffic Based On Chaos Theory With Support Vector Machine

Nowadays Computer Networks have been the most important way that people communicate information with each other. Along with the rapid development of Network, increasingly complicated requirements in practice make Network management much more difficult. Meanwhile, congestion, security and malfunction in Networks actually are handicapping the normal developing of information based society. How to effectively utilize Networks' resources, protect Networks' security, adjust and upgrade Networks' equipments in time is currently the hotspot in research.Recently, as a result of long progress of Network traffic forecasting, it has been a type of key technology to cope with difficulties mentioned above. Researchers introduced traffic forecasting based strategy to dynamically allocate resource according to different types of operation in QoS mechanism; besides, it also plays significant role in other aspects such as Network security, Network layout etc., for instance, traffic forecasting based IDS (Intrusion Detection System) and optimizing scheme of Network's bandwidth would be implemented.Through wide study and research within traditional Time Series Analysis, Chaotic Time Series Analysis, and computational intelligence based forecasting algorithms, short-term forecasting algorithm of Networks' traffic relying on Chaos theory and Support Vector Machine is investigated in profundity, and then a novel LSVM-DTW-K algorithm with better forecasted preciseness and stronger adaptability is proposed. The main contributions of this thesis are: putting forward a Chaotic Time Series Analysis and Support Vector Machine based short-term Networks' traffic forecasting algorithm which is adept at scenarios that only small scale dataset is available. With the previously proposed Local Support Vector Machine forecasting algorithm, after analyzing and considering the vulnerabilities in modeling, using Dynamic Time Wrapping to replace Euclid's Distance in measuring the similarity between a pair of vector is introduced eventually; Furthermore, the "dynamic K" strategy is designed to dynamically organize those reasonable ones from nearest neighboring points in forecasting, instead of keeping the number of nearest neighboring points invariable all along, which brings about the performance stronger in adaptability as well as in preciseness. A set of experiment employed two datasets respectively from campus Wired Network's IP traffic and campus Wireless Network's IP traffic demonstrate proposed LSVM-DTW-K algorithm certainly can perform better than Local Support Vector Machine algorithm not only in the forecasted preciseness, but in adaptability as well.