IX TELEKOMUNIKACIONI FORUM TELFOR'2001, Beograd, 20-22.11.2001.god. NEURAL NETWORKS BASED MODEL OF A HIGHLY NONLINEAR PROCESS

In this paper two problems will be analyzed: First, the obtain the more accurate prediction than using traditional linear or nonlinear regression approach. For this purpose we shall use the tree layers NN based on Levenberg-Marquardt algorithm. Second, it is known that recurrent neural networks (RNN) are very useful in nonlinear process modeling. Consequently, the trend, i.e. a line of general direction of considered nonlinear process, will be extracted by the Elman recurrent network, NN Toolbox, [1] .The presented data (Fig.2) illustrate periodical oscillations in the considered process.