A neural approach to inverse problem of teletraffic systems

For analyzing and evaluating the performance of communication systems, beforehand, it is necessary to know the traffic parameters for describing the system (traffic model or queueing model) and identifying its components (interarrival and service processes, and the number of servers and buffers). However, connecting a lot of computers and local area networks (for example, the Internet), makes the traffic complicated. As a result, specifying the traffic parameters becomes very difficult. Therefore, we must find a new method which specifies what kind of traffic parameters are used to analyze and evaluate the performance of the systems. We propose a method using a simple three layer neural network to solve the inverse problem of the traffic system. The inverse problem is the problem that specifies the traffic parameters (input) from the performance indices (output) of the traffic system. The algorithm of this method consists of the learning phase and the specifying and classifying phase. In the learning phase, the weight values of the neural network are determined by using the back propagation algorithm. For training of the neural network, we use the performance indices of the traffic system as the learning data, and the corresponding traffic parameters of the system as the teaching signal. In the specifying and classifying phase, using the trained neural network, we can classify a given traffic system. The proposed method provides a good accuracy for the specification of the parameters of unlearned traffic systems too.