АНАЛІЗ НЕЙРОМЕРЕЖЕВИХ МОДЕЛЕЙ В ЗАДАЧАХ ОПТИМІЗАЦІЇ ТЕХНОЛОГІЇ ЕНЕРГОКОНДЕНСОВАНИХ СИСТЕМ
Journal Title: Математичне моделювання - Year 2018, Vol 1, Issue 1
Abstract
ANALYSIS OF THE NEURAL NETWORK MODELS IN THE PROBLEMS OF OPTIMIZING THE TECHNOLOGY OF ENERGY CONDENSED SYSTEMS Korotka L.I. Abstract Optimizing the technology of energycondensed systems is the subject area in this article. In the technology of explosives without TNT, one of the problems is the determination of technological parameters. There are results of full-scale experiments, obtained at a private joint-stock company in Zaporozhe. The formulation of the problem is to determine the optimal technological parameters when using new technologies of nitrate condensed systems. The mathematical formulation of the problem consists in determining the output parameter - the strength of ammonium nitrate with the available input data From a formal point of view, the formulated problem is an approximation problem. The author knows the classical approaches to solving problems of this class, but it is proposed to use the mathematical apparat of the theory of artificial neural networks. Based on the theorem of Hecht-Nielsen and Kolmogorv-Arnold and the consequences of them, it is proposed to consider one and two-layer neural networks. The technology of creating a neural-network component is considered in stages. The preliminary stage of preparation of training samples for a neural network is disassembled. In the event that the specified network error is sufficiently small, a separate software module is developed to obtain the required volume of training samples. The parameters of the choice of network architecture and its learning are analyzed. Two types of sigmoidal activation functions are considered. The process of learning the neuro-network component and its features is described. As an algorithm, an algorithm for back propagation of an error without an inertial component is used. The trained neural networks were tested on the reserved samples. An analysis of numerical results is made, on the basis of which it is possible to choose from the set of applicants the best neural network component. Of the many applicants with almost identical parameters, it was decided to choose a neural network with the simplest architecture. The trained network was further built into the overall process of optimizing the technology of energy condensed systems. References [1] Kovalenko Y.L., Kyiashchenko D.V. Tehnologiya modifitsirovaniya agrarnoy ammiachnoy selitryi v proizvodstve energokondensirovannyih sistem. Science and education a new dimension. Natural and technical sciences. ¬ 2015. ¬ III (8), Issue 73. ¬ pp. 107 ¬110 (in Russian). [2] Kuprin V.P., Kuprin O.V., Ishchenko M.I., Savchenko M.V., Kovalenko I.L. Sposib vyhotovlennia hranulovanykh vybukhovykh su-mishei dlia pnevmatychnoho zariadzhannia iz amiachnoi selitry i ridkoho palyva [Method for manufacturing granular explosive mixtures for pneumatic charging from ammonium nitrate and liquid fuel]. Patent 106118 Ukraina, MPK S06V31/28, F42D3/04. № u 2015 13112; zaiavl. 30.12.2015; opubl. 11.04.2016, Biul. № 7. [3] Kovalenko І., Stupnik N., Korolenko M., Kiyaschenko D., Batareev A. “Cartridged and granulated explosive substances of grade Ukrainit for underground mines” Metallurgical and mining industry. – 2016. ¬ №8. ¬ pp. 59¬64. Access to the magazine: http://www.metaljournal.com.ua/MMI-2016-No-8. [4] Kovalenko I.L., Kuprin V.P. Vplyv poperednoi pidhotovky na vbyraiuchu zdatnist i mitsnist hranul amiachnoi selit-ry marky B [Effect of preliminary preparation on the absorbency and strength of granules of ammonium nitrate of grade B]. Naukovi visti NTUU «KPI» – Scientific reports of NTUU “KPI“, 2014, № 6 (98), pp. 110–114 (in Ukrainian). [5] Korotkaya L.I. Ispolzovanie neyronnyih setey pri chislennom reshenii nekotoryih sistem differentsialnyih uravneniy [The use of neural networks in the numerical solution of certain systems of differential equations]. Vostochno-evropeyskiy zhurnal peredovyih tehnologiy. Vostochno-evropeyskiy zhurnal peredovyih tehnologi. – Eastern European Journal of Advanced Technologies, 2011, № 3/4 (51), pp. 24 ¬ 27 (in Russian). [6] Korotka L.I. Funkcional`na pidsy`stema racional`-nogo vy`boru arxitektury` nejronnoyi merezhi [Functional subsystem of the rational choice of the architecture of the neural network]. Visny`k Xersons`kogo nacional`nogo texnichnogo universy`tetu ¬ Bulletin of Kherson National Technical University, 3(62), Tom I. (Fundamenta-l`ni nauky`- Fundamental Sciences), 2017, pp. 55–59 (in Ukrainian). [7] Kruglov V.V., Dli M.I., Golunov R.YU. Nechetkaya logika i iskusstvennye nejronnye seti [Fuzzy logic and artificial neural networks], Moskva: Fizmatlit, 2001. − 224 p. [8] Pyatkovskiy O.I. Intellektualnyie informatsion-nyie sistemyi. (Neyronnyie seti). Uchebnoe posobie [Intelligent information systems. (Neural networks)] / Alt. gos. tehn. Un-t im. I.I. Polzunova. Barnaul: Izd-vo AltGTU, 2010. – 125 p. [9] Haykin S. Neyronnyie seti: polnyiy kurs [Neural networks: full course], 2-e izda-nie. per. s anrl. / Saymon Haykin. – M.: Izdatelskiy dom "Vilyams", 2006. – 1104 p.
Authors and Affiliations
Л. І. Коротка
Keywords
Related Articles
- EP ID EP294908
- DOI 10.31319/2519-8106.1(38)2018.129020
- Views 76
- Downloads 0
Математическая модель плавления утяжеленного комбинированного алюмосодержащего раскислителя цилиндрической формы в защитной оболочке
This article presents the one- and two-dimensional mathematical model of melting aluminum-weighted combined deoxidizer cylindrical shape in the containment in the melt and asymmetric boundary conditions at the interface...
О выборе параметров ЕМ-алгоритма для разделения смеси распределений
ON THE CHOICE OF THE PARAMETERS OF THE EM ALGORITHM FOR THE SEPARATION OF THE MIXTURE OF DISTRIBUTIONS. Shumeiko A.A, Iskandarova-Mala A. Abstract The paper considers one of the types of non-hierarchical clustering - th...
Окремі випадки застосування методів чисельного ресамплінгу в прикладних задачах
In this paper we investigated methods of numerical resampling, which combines three different approaches, which differ according to the algorithm, but similar in essence: randomization (or permutation), bootstrap and jac...
. Computer methods for compiling an entry of explanatory combinatorial dictionary belonging to "meaning↔text" theory within the task of text automatic generation
КОМП’ЮТЕРНА МЕТОДИКА ФОРМУВАННЯ СТАТТІ ТЛУМАЧНО- КОМБІНАТОРНОГО СЛОВНИКА ТЕОРІЇ «СМИСЛ↔ТЕКСТ» В РАМКАХ ЗАВДАННЯ АВТОМАТИЧНОЇ ГЕНЕРАЦІЇ ТЕКСТІВ Ковилін Є.Р., Волковський О.С. Реферат Мета. Розробка комп’ютерної моделі то...
Оптимальные непересекающиеся фильтры в обработке изображений
The article suggests an image processing by iterative filtering scheme using superposition disjoint filters.