Radiotekhnika
Publishing house Radiotekhnika

"Publishing house Radiotekhnika":
scientific and technical literature.
Books and journals of publishing houses: IPRZHR, RS-PRESS, SCIENCE-PRESS


Тел.: +7 (495) 625-9241

 

Implementation of complementary correction in automatic control systems for trajectorial movements of technological objects using a neural network controller

Keywords:

A.A. Kobzev – Dr.Sc.(Eng.), Professor, Head of Department of Mechatronics and Electronic Systems of Vehicle, Vladimir State University named after A.&N. Stoletovs
E-mail: kobzev42@mail.ru
Yu.M. Monakhov – Ph.D.(Eng.), Associate Professor, Department of Informatics and Information Security,
Vladimir State University named after A.&N. Stoletovs
E-mail: unklefck@gmail.com
A.V. Lekareva – Post-graduate Student, Department of Mechatronics and Electronic Systems of Vehicle,
Vladimir State University named after A.&N. Stoletovs
E-mail: tasya671@rambler.ru


The article considers dynamic correction in automatic control systems of trajectorial movement of technological objects implemented by artificial intelligence methods. The analysis of the structures of ACS with complementary correction allows us to propose two schemes for the inclusion of a neural network controller in the ACS circuit: 1) the generation of an autonomous additional component by the neural network controller sent towards the control action, and the formation of a channel for this component; 2) software control correction in the device for generating the control action. These structures assume dynamic learning of the neural network. As an error signal used to adjust the weights of a neural network, authors suggest to use an error signal between the output coordinate of the reference model and the control object. Here, the corrections to the control law, generated in the system correction circuit, are assumed to be unknown, i.e. there is no reference value of the output signal of the neural network.
The work outlines approaches to choosing the number of layers, neurons in them, activation functions and the delay time of input signals of the network. A modification of the gradient descent method for neural network learning is proposed.
The results of the study of the correction contour with the neural network regulator confirm the operability of the proposed approach. However, when the parameters of perturbations are changed, an increase in the error in the working of the master signal is observed at constant values of the neural network parameters. Reducing the time of discreteness of the neural network significantly reduces the misalignment of the output values of the control object and the reference model. Therefore, it is advisable to indirectly determine the perturbation parameters based on an error proportional to this effect to develop an algorithm for determining optimal parameters and the time of the neural network discreteness.

References:
  1. Vasil'ev V.I., Il'yasov B.G. Intellektual'ny'e sistemy' upravleniya. Teoriya i praktika: Ucheb. posobie. M.: Radiotexnika. 2009. 392 s. ISBN 978-5-88070-225 1.
  2. Novoselov B.V., Goroxov Yu.S., Kobzev A.A., Shhitov A.I. Avtomaty'-nastrojshhiki sledyashhix sistem / Pod red. B.G. Novoselova. M.: E'nergiya. 1975. 264 s.
  3. Lekareva A.V. Obespechenie invariantnosti oshibki po vozmushheniyu v sistemax avtomaticheskogo upravleniya traektorny'mi peremeshheniyami texnologicheskix ob''ektov // Nauchno-texnich. vestnik informaczionny'x texnologij, mexaniki i optiki. 2016. T. 16. № 5. S. 787−795.
  4. Kobzev A.A., Lekareva A.V., Maxfuz A.A. Algoritmy' upravleniya texnologicheskim manipulyatorom gidrorezaniya nefteprovodov s uchetom speczifiki vy'polnyaemogo proczessa // Dinamika slozhny'x sistem. 2016. S. 34−40.
  5. Gorban A.N., Wunsch D. The general approximation theorem // Proceedings of the IJCNN. Anchorage: IEEE. 1998. P. 1271−1274.
  6. Stone M.N. The generalized Weierstrass approximation theorem // Math. Mag. 1948. V. 21. P. 167−183, 237−254.
  7. Sheela K.G., Deepa S.N. Review on methods to fix number of hidden neurons in neural networks // Mathematical Problems in Engineering. 2013. V. 2013. P. 1−11.
  8. Bartlett E.B. Dynamic node architecture learning: An information theoretic approach // Neural Networks. 1994. № 7. P. 129−140.
  9. Hecht-Nielsen R. Kolmogorov's Mapping Neural Network Existence Theorem // Proc. Int. Conf. Neural Networks. New York: IEEE Press. 1987. Vol. III. P. 11−14.
  10. Kůrková V. Kolmogorov's theorem is relevant // Neural Computation. 1991. V. 3. № 4. P. 617−622.
  11. Xajkin S. Nejronny'e seti: polny'j kurs: Per. s angl. Izd. 2-e. M.: OOO «I.D. Vil'yams». 2017. 1104 s.
  12. Kim S.-S. Time-delay recurrent neural network for temporal correlations and prediction // Neurocomputing. 1998. V. 20. P. 253−263.
  13. Eremenko Yu.I., Glushhenko A.I. O razrabotke metoda vy'bora struktury' nejronnoj seti dlya resheniya zadachi adaptaczii parametrov linejny'x regulyatorov // Sb. trudov «Upravlenie bol'shimi sistemami». 2016. № 62. S. 75−123. ISSN 1819-2440.

© Издательство «РАДИОТЕХНИКА», 2004-2017            Тел.: (495) 625-9241                   Designed by [SWAP]Studio