Clustering Analysis of Stochastic Node Networks

Abstract

In this paper is solving the problem of performance evaluation of discrete Markov chains in which the topology is unknown in advance. Two heuristic classification procedures for clustering of classes are presented. The first is realized with the help of self-organizing Kohonen neural network in which the target classes are describing with winner weights vectors. The efficiently performance of the clustering topological problem at the second procedure uses the combination of minimum variance with rare event cross entropy method. As an illustration, are shown numerical examples for two-dimensional stochastic node network, which describes the feasible states of Jackson tandem queue.