Self-Similar Traffic Prediction Algorithm Based on an Improved Kalman Filtering

Abstract

In recent years, network congestion is more intractable than ever before. More often than not, congestion results from excessive traffic packets at network nodes. By means of the correlation lying in traffic flow, traffic can be predicted to mitigate congestion. In this paper, a traffic prediction algorithm based on an improved Kalman filtering is proposed. Independent of the feedback information from traffic sources, current and former traffic measured in nodes are used to predict the traffic load in the next time. In addition, the state equation and the observation equation are formulated. As the noise statistics of state equation and observation equation are undetermined, an on-line estimation method with forgetting factor is used to estimate noise statistics. This algorithm has low space and time complexity. Simulation results show that, compared with other algorithms, the proposed algorithm offers more accurate prediction of network traffic.