A proposed intelligent bandwidth management system based on Turksen's Fuzzy Function approach using reinforcement learning forecasting.

摘要

In the telecom industry, there is a continuously increasing pressure to find more effective ways of managing available resources, such as bandwidth. The nature of the bandwidth allocation problem lends itself to the field of industrial engineering and operations research, because it is an optimization problem. However, upon further inspection of the complexity of this problem, it eventually becomes clear that finding an optimal solution will not be computationally feasible. However, by employing appropriate computational search techniques, a good solution may be approximated.; This thesis approaches this problem in a systematic, two-stage process. The first is to use Reinforcement Learning (RL) techniques to learn the dynamic nature of the environment. The resulting forecasts shall then be used as an input for a newly proposed fuzzy system modeling approach, namely, Turksen's Fuzzy Functions [17]. A model comparison between ordinary regression and Fuzzy Functions is then made. This thesis purports that the Fuzzy Function approach represents a paradigmatic advancement from classical regression. This thesis may be of interest to industrial engineers involved in planning and decision support, computer scientists studying artificial intelligence, and engineers and managers in the telecom industry.