In this work, we propose an adaptive predictive flow control scheme based on the TSK fuzzy model for congestion control in broadband networks. The proposed control scheme intends to avoid congestion by applying a TSK type model to predict the buffer queue length. For this end, we developed an adaptive training algorithm for the TSK model that was incorporated into the proposed control scheme, achieving low loss rate network performance. Our developed fuzzy predictor consists of a two-step algorithm: an adaptive training stage with covariance resetting and a gradiente-based learning algorithm for refining the previous part of the prediction procedure. An evaluation of the proposed predictor is carried out by using real network traffic traces. As essential part of the proposed congestion control scheme, we derive an analytical expression involving the fuzzy model parameters for the control of the flow rates that minimizes the queueing length variance. Further, a network environment is considered and the congestion control is applied to each node of the analyzed scenery. Simulations are made to compare the performance of the proposed adaptive control algorithm to those of some existing control approaches in literature.
Flow Control; Adaptive Prediction; Fuzzy Modeling; Network Traffic
