Publication detail

The genetic algorithm used for the optimization of weights and biases of neural networks

HAMMER, M.

Czech title

Genetický algoritmus pro optimalizaci vah neuronové sítě.

English title

The genetic algorithm used for the optimization of weights and biases of neural networks

Type

conference paper

Language

en

Original abstract

The life of the insulating systems of electric rotary machines is strongly dependent upon electrical and thermal features of the insulating material used. The subject of the diagnostics is to specify the condition of insulation used. At present days, the most popular diagnostic tools are the methods of artificial intelligence, and one method is a neural network. However, this tool has many variable parameters, and the resulting effect is dependent upon its suitable setting. Weights and biases that connect neurones in each layer of the neural network are ranked among these parameters. This paper is concentrated on the use of genetic algorithms for the optimization of weights and biases of the neural network that is used as a diagnostic tool for winding insulation of electric rotary machines. In this case, the optimization of weights and biases means that the neural network exhibits the minimum absolute mean errors for the diagnostics of winding insulation. The paper describes the architecture and the setting of the neural network, the optimization method, the genetic algorithm and the input/output data. The optimization curves are shown in the graphs, and the results achieved after the optimization are shown in tables. The use of the diagnostic tool to solve the problems investigated is assessed in tables. The optimization method, the genetic algorithm and neural networks were programmed in Matlab 6 environment. Also, all simulations and the calculated values were obtained by means of this product.

Czech abstract

Článek se zabývá diagnostikou izolačního materiálu elektrických strojů.

English abstract

The life of the insulating systems of electric rotary machines is strongly dependent upon electrical and thermal features of the insulating material used. The subject of the diagnostics is to specify the condition of insulation used. At present days, the most popular diagnostic tools are the methods of artificial intelligence, and one method is a neural network. However, this tool has many variable parameters, and the resulting effect is dependent upon its suitable setting. Weights and biases that connect neurones in each layer of the neural network are ranked among these parameters. This paper is concentrated on the use of genetic algorithms for the optimization of weights and biases of the neural network that is used as a diagnostic tool for winding insulation of electric rotary machines. In this case, the optimization of weights and biases means that the neural network exhibits the minimum absolute mean errors for the diagnostics of winding insulation. The paper describes the architecture and the setting of the neural network, the optimization method, the genetic algorithm and the input/output data. The optimization curves are shown in the graphs, and the results achieved after the optimization are shown in tables. The use of the diagnostic tool to solve the problems investigated is assessed in tables. The optimization method, the genetic algorithm and neural networks were programmed in Matlab 6 environment. Also, all simulations and the calculated values were obtained by means of this product.

Keywords in English

Genetic algorithm, System optimization, Neural network, Artificial intelligence

Released

01.01.2002

Publisher

S.C.ICPE

Location

Constanta, Rumunsko

ISBN

973-95041-3-2

Book

Study and control of corrosion in the perspective of sustainable development of urban distribution grids

Pages count

5

BIBTEX


@inproceedings{BUT5355,
  author="Miloš {Hammer},
  title="The genetic algorithm used for the optimization of weights and biases of neural networks",
  booktitle="Study and control of corrosion in the perspective of sustainable development of urban distribution grids",
  year="2002",
  month="January",
  publisher="S.C.ICPE",
  address="Constanta, Rumunsko",
  isbn="973-95041-3-2"
}