P. Seidl, I. Taufer

A short-circuited slot line has been investigated with the aim to construct its model in the form of the artificial neural network (ANN). Mathematical description of this line is complicated function with dependence of source signal and physical properties of the line to the terminal impedance. The aim was to get the impedance dependence on slot width and source signal frequency. The model represents the system with two inputs - the slot width and the source signal frequency - and one output - the real or imaginary component of the line terminal impedance. The «Neural Network Toolbox» of the computing system MATLAB/SIMULINK was used to create artificial neural network with two inputs, one output and one hidden layer. Proposed neural network was trained by the Backpropagation algorithm and its modification, the Levenberg-Marquardt's method. Trained network described very well real behaviour of investigated line. Average error between experimental impedance and model data was 1.22 %. The results demonstrated the capability of using of the ANN as the tool for the modelling of the investigated system

1. Macháč J, Zehentner J., Menzel W. Short and open circuited slot line. In Proceedings International Microwave and Optoelectronics Conference. Rio de Janeiro (Brasil), July 24-27, 1995. Rio de Janeiro: Brazilian Microwave and Optoelectronics Society. 1995. P. 856 - 861. ISBN 0-7803-2674-1.
2. Fausset L.Fundamentals of Neural Network: Architectures, Algorithm and Applications. New Persey: Prentice Hall, Inc. 1994.
3. Masters T. Practical Neural Network Recipes in C++. Boston: Academia Press, Inc. 1993. ISBN 0-12-479040-2.
4. Leondes T.C. Industrial and Manufacturing Systems. San Diego, California, USA: Academia Press, Inc. 1989.
5. ISO/IEC 2382-34:1999 Information technology - Vocabulary - Part 34: Artificial intelligence - Neural network.
6. Demuth H., Beale M. Neural Network Toolbox for Use with MATLAB. Natick (USA): MathWorks, Inc., 1994.