Description: The method of intellectual identification and prediction of traffic in information telecommunication networks is proposed. It consists of the definition of dimensions of traffic models, its identification and prediction using artificial intelligence systems, choice of block-oriented structures of models and the composition of global and local optimization methods. Determination of dimensions of models is done by calculation of correlation entropy, correlation interval of predictability, correlation dimension of the attractor, the dimension of attachment to the attractor. For global optimization methods applied genetic algorithms, multiobjective optimization, direct search, simulated annealing and threshold acceptance. Models of Wiener, Hammerstein, and their compositions are used as block-oriented models. As the basis functions are used cascaded feedforward neural network, neural network with radial basis functions and hybrid network with fuzzy logic. For parametric optimization was used the criterion of regularity, calculated on the test sample, and for the global –the criterion of minimum bias, based on the analysis of solutions. By modeling, the efficiency of using the proposed method is estimated using the example of identification and prediction of experimental data - traffic transmitted via the Internet.
Keywords: identification, prediction, traffic, neural networks, fuzzy logic, model structures, global optimization