Prediction of Queue Lengths in the Multi-Queue-Single-Processor Queuing System Based on Fuzzy-Neural Approach

In this paper the problem of prediction of queue lengths in the multi-queue-single-processor queuing system is considered. Efficient solutions for this problem are crucial in many technical appliances, such as load balancing in multi-processor systems. Here, a fuzzy-neural model for approximation of the average future system load along with model identification algorithm are proposed. Moreover, condition for convergence of learning is given. The quality of the model is evaluated by means of computer simulation. In order to quantitatively assess the performance of the model, it is compared to other existing prediction models: linear and nonlinear perceptron, Takagi-Sugeno model, recurent neural network and moving average filter. The accuracy of prediction of considered models is compared with respect to the normalized average prediction error criterion. In the simulation real-world data is used as models input and output series. The data concern characteristics of traffic flowing through a network router interconnecting two clouds of network clients. The task of prediction models is to predict average future load of parallel processors in the multiprocessor router. It is shown, that the performance of the proposed fuzzy-neural approach outperforms other ones for various scenarios of network setup.