Application of artificial neural network for solving problems of forecasting the movement of ground objects

DOI: 10.34759/trd-2022-123-17

Аuthors

Sokolov D. Y.

National Helicopter Center Mil & Kamov, 26/1, Garshina str., Tomilino, Moscow region, 140070, Russia

e-mail: dsokolov@ mi-helicoptr.ru

Abstract

The article deals with considering the artificial neural network application for solving the problems of the ground-based object motion predicting. It proposes a new approach to the neural network architecture elaboration and applying its training techniques as applied to the specifics of the problem of motion trajectory prediction in the space of parameters. Estimation of various training techniques is given, and optimal parameters of coefficients, employed in the neural network training algorithms are determined.

The task of the ground-based objects motion predicting is up-to-date as applied to both combat helicopters actions, in the case of their aiming at a moving maneuvering target, and civil aviation, including unmanned aircraft, when it is necessary to track various ground objects.

Application of the known types of neural networks for solving the problem of the motion predicting of a ground object (as an image of the object trajectory en masse) does not account for the sequence of passing the trajectory points by the object, which is an important information for solving the problem of the object motion predicting. Additional data processing and mathematical transforms allow solving the problem of accounting for the information on the sequence of passing points. However, this increases the overall complexity of calculations and may level the effectiveness of NN application for solving the forecasting problem.

The article proposes a certain architecture for a neural network building, in which each neuron corresponds to a separate point or region of space where the movement of an object being performed. This architecture assumes elaboration of a recurrent neural network, which imposes certain specifics on the training process and subsequent application of the neural network. Thus, the training process of a recurrent neural network should be iterative until a certain minimum of the objective function is reached.

The article discusses various ways of a neural network training and determines the most suitable ones from the standpoint of minimizing the errors and duration of the training cycles.

This article considers the movement in two coordinates (on a plane). However, in the general case, the space of parameters, in which the movement occurs, may contain greater number of coordinates (hyperplane). Both training and application of predictive neural network for the motion prediction in the space of parameters with any number of coordinates will be realized by the technique similar to the one for the motion on the plane.

In the prospect, the proposed version of a neural network building may be applied to predict the movement of various objects, including the abstract ones in a multidimensional hyperspace of their parameters.

Keywords:

neural networks, motion trajectory, object motion prediction, parameter space, ground object, neural network architecture, neural network training algorithm

References

1. Vasil'ev V.A., Fedyunin P.A., Danilin M.A., Vasil'ev A.V. Trudy MAI, 2019, no. 105. URL: http://trudymai.ru/eng/published.php?ID=104241

2. Lebedev G.N., Mirzoyan L.A., Efimov A.V. Trudy MAI, 2011, no. 48. URL: http://trudymai.ru/eng/published.php?ID=27150

3. Goncharenko V.I., Zheltov S.Yu., Knyaz' V.A., Lebedev G.N., Mikhailin D.A., Tsareva O.Yu. Izvestiya Rossiiskoi akademii nauk. Teoriya i sistemy upravleniya, 2021, no. 3, pp. 39-56. DOI: 10.31857/S0002338821030057

4. Bel'skii A.B. VI Vserossiiskaya voenno-nauchnaya konferentsii «Aktual'nye problemy vooruzhennoi bor'by v vozdushno-kosmicheskoi sfere»: sbornik trudov, Voronezh, VUNTs VVS VVA, 2020, pp. 66-75.

5. Lebedev G.N., Mirzoyan L.A., Efimov A.V. Mekhatronika, avtomatizatsiya, upravlenie, 2009, no. 11, pp. 60 – 65.

6. Anan'ev A.V., Rybalko A.G., Ivannikov K.S., Klevtsov R.P. Trudy MAI, 2020, no 115. URL: https://trudymai.ru/eng/published.php?ID=119975. DOI: 10.34759/trd-2020-115-18

7. Ivashova N.D., Mikhailin D.A., Chernyakova M.E., Shanygin S.V. Trudy MAI, 2019, no. 104. URL: http://trudymai.ru/eng/published.php?ID=102223

8. Anan'ev A.V., Rybalko A.G., Ivannikov K.S., Klevtsov R.P. Trudy MAI, 2020, no. 115. URL: https://trudymai.eng/ru/published.php?ID=119975. DOI: 10.34759/trd-2020-115-18

9. Sokolov D.Yu. Vooruzhenie i ekonomika, 2012, no. 1 (17), pp. 50-57.

10. Vasil'eva V.I., Il'yasova B.G., Kusimova S.T. Neirokomp'yutery v aviatsii (Neurocomputers in aviation), Moscow, Radiotekhnika, 2004, pp. 120-180.

11. Chebatko M.I. Zarubezhnaya elektronika, 1994, no. 11-12, pp 40-44.

12. Osovskii S. Neironnye seti dlya obrabotki informatsii (Neural networks for information processing), Moscow, Finansy i statistika, 2004, 343 p.

13. Lebedev G.N., Rumakina A.V. Aviakosmicheskoe priborostroenie, 2014, no. 5. pp. 3–8.

14. Galyutin V.B. Aviakosmicheskoe priborostroenie, 2002, no. 3, pp. 29-34.

15. Venttsel' E.S. Ovcharov L.A. Teoriya sluchainykh protsessov i ee inzhenernye prilozheniya (Theory of random processes and its engineering applications), Moscow, Vysshaya shkola, 1998, 354 p.

16. Bel'skii A.B., Sakharov N.A. Aviakosmicheskaya tekhnika i tekhnologiya, 2013, no. 1 (69), pp. 43-46.

17. Anan'ev A.V., Rybalko A.G., Lozarak A.V. Trudy MAI, 2020, no. 112. URL: https://trudymai.ru/eng/published.php?ID=116584. DOI: 10.34759/trd-2020-112-19

18. Bel'skii A.B. VII Mezhdunarodnaya nauchno-prakticheskaya konferentsiya «Aktual'nye voprosy issledovanii v avionike: teoriya, obsluzhivanie, razrabotki» («AVIATOR»): sbornik statei. Voronezh: VUNTs VVS VVA, 2020, pp. 37-46.

19. Bel'skii A.B. Aviakosmicheskaya tekhnika i tekhnologiya, 2018, no. 1,2 (85,86), pp. 67-74.

20. Andreev D.V. Trudy MAI, 2018, no. 101. URL: https://trudymai.ru/eng/published.php?ID=96651

Download