Design of a network traffic prediction model using the Kalman filter

Thumbnail Image
Journal Title
Journal ISSN
Volume Title
Network traffic prediction is of immense interest to industry since it supports decision making in management and control, and eventually user satisfaction. In this project, a Kalman filter-based model was developed for predicting network traffic. The focus was on telecommunication, transportation and computer networks. Traffic volume per unit time, or traffic flow rate, observed in a particular time interval, was utilized to predict the traffic flow rate for the next time interval by recursively computing relevant parameters of incoming traffic data. Relevant parameters include the process and measurement noise covariances, the Kalman filter gain and the a-priori and a-posteriori state and covariance estimates. The model makes use of the Kalman filter to carry out the prediction, and was tested using traffic sets with low, average and high autocorrelation. By means of a LabVIEW VI (simulation tool) different parameters were varied and their effects on the prediction model observed. LabVIEW was employed for its superior simulation features, with an integrated MATLAB block for optimization. Working with a 20% error tolerance, prediction accuracies approached 90%, and this process yielded an improved short-term traffic flow rate prediction model. The model carried out prediction for a single time-step ahead but, with refinements or modifications, it may be employed for multi-step prediction, converting it to a long-term predictor.
A thesis submitted to the Department of Telecommunications Engineering. Kwame Nkrumah University of Science and Technology, in partial fulfilment of the requirement for the degree of Master of Science