Neural Network Adaptive Control with Smith Predictor for Flow Rate in Networks
Shao Hui-he · Acta Simulata Systematica Sinica · 2003
The network time delay has a great adverse effect on the rate-based ABR flow control in ATM networks. Smith predictor is an effective means that overcomes the large time delay. However, it is very sensitive to model error, so it is not satisfactory to use only Smith predictor in real networks which has indeterminacy of the time delay. The paper designs the neural network adaptive controller with Smith predictor for the flow control, which can overcome the adverse effect caused by the time delay and its indeterminacy well. Thus the source rates can respond to the changes of network status rapidly. Compared with PID Smith predictor control, this scheme has much better adaptability and robustness which are applicable to actual networks, and much lower buffer capacity which is necessary for no cell overflow and link bandwidth full utilization.