Dušan Petrovački

Original Contribution: Evolutional development of a multilevel neural network

The paper describes changes in the structure (anatomy) of a multilevel neural network which is trained by back-propagation algorithm. A structural change is both a random process and an environment-dependent process. The mechanisms of cell propagation and degeneration of inactive synapses and inactive cells are considered. The change in the neural network structure occurs during the training and running of the system, which enables dynamic adaptation of the network capacity to the complex problem of the network interaction with the environment. Structural changes in a multilevel neural network are tested on several tasks of the network training.

Optimal control of a heat conduction problem

SummaryThis paper treats the problem of optimal control of a class of linear distributed parameter systems. Fourier series are used for transformation of a one-dimensional heat conduction equation into a system of ordinary differential equations. Necessary conditions for the optimal control are derived by means of Pontryagins maximum principle. The application of obtained results is shown through two examples. The method in this paper is limited to linear systems.ZusammenfassungDie Arbeit befaßt sich mit dem Problem der optimalen Steuerung einer Klasse von Systemen deren mathematisches Modell durch partielle Differentialgleichungen gegeben wird. Fouriersche Reihen werden benutzt, um die eindimensionale Wärmeleitungsgleichung in das System gewöhnlicher Differentialgleichungen zu transformieren. Die Bedingungen zur optimalen Steuerung wurden unter Benutzung des Maximumprinzips von Pontrjagin abgeleitet. Die Anwendung der gewonnenen Ergebnisse wird mit Hilfe zweier Beispiele dargestellt. Die Methode, die in der Arbeit dargelegt wird, ist auf lineare Systeme begrenzt.

Land Use Management Based on GPS Technology, and Satellite and Ground Remote Sensing Technologies

The paper presents an overview of the GPS technology, satellite remote sensing technologies, ground remote sensing technologies, and GIS technologies applications to the land use management. The technological method, developed by the Centre for Geo-Information Technologies and Systems Novi Sad, aimed for but not restricted to agricultural land use management using GPS, satellite remote sensing, geo-radar sensing and GIS technologies is presented as a case study.

Bridging of layers of neural networks

This paper involves the evolution of neural networks based on parameters gained from a backpropagation training algorithm. The bridging of layers of neural networks is suggested as a new process in this evolutionary process, in addition to cell division and degeneration of synapses and cells. The process of bridging the layers was performed in three different ways in order to find an appropriate algorithm.

Predictive control algorithms for MIMO systems

Generalized predictive controller and generalized minimum variance control algorithms for multi-input-multi-output (MIMO) systems are presented. Both algorithms can be applied to any MIMO system regardless of the number of inputs and outputs. Simulation results are presented.