Dedicated to the blessed memory
of a talented young man,
a brave patriot of his Motherland EDUARD VALERIEVICH KONONCHUK
(11/15/1997 - 04/22/2022).
Section 1
Math modeling
E. V. Kononchuk, T. V. Ermolenko, T. O. Shishunov. MACHINE LEARNING MODELS FOR ASSESSING THE PROBABILITY OF ACCIDENT AND ITS SEVERITY.
UDC 004.8:004.6 Annotation:The article presents the results of the analysis of the significance of factors affecting the occurrence of
road accidents, as well as an analysis of the effectiveness of predictive models built on the basis of
decision trees and neural networks. The models were trained on a dataset of road traffic accidents in
America taken from the Kaggle website. Keywords: exploratory data analysis, random forest, principal component analysis, multilayer perceptron.
V. G. Chernikov. DEVELOPMENT OF A METHODOLOGY FOR DETERMINING THE PARAMETERS OF AN ADAPTIVE DISCRETE CONTROLLER OF THE WIND WHEEL ROTATION SPEED .
UDC 51-7:004.052:622.53 Annotation:The article proposes a method for determining the parameters of a discrete speed controller of a
wind wheel, taking into account its mathematical model. The possibility of using a neural network to
determine the nonlinear parameters of a wind wheel as an object of regulation is considered. The
duration of the control program, which implements the function of an adaptive discrete speed
controller based on an industrial programmable logic controller, is determined. Keywords: wind turbine, control loop, discrete controller.
O. A. Shevchuk. MATHEMATICAL MODELING OF THE STRESS-STRAIN STATE OF STEEL VERTICAL CYLINDRICAL TANKS.
UDC 004.94 Annotation:The article proposes a mathematical model for calculating the stress-strain state of a tank for
storing petroleum products with walls of constant thickness that are exposed to the internal
pressure of the liquid, made by approximating the numerical solution of the differential equation by
geometric modeling methods. Comparison of the obtained results with the reference solution
demonstrates a high degree of reliability of the proposed method. The implemented approach of
mathematical modeling is a more universal tool in relation to changing both the initial and boundary
conditions of the problem, and the differential equation itself. Keywords: Koch's curve, Dragon outline, Fern leaf, reflection, rotation, symmetry.numerical modeling, differential equation, geometric interpolant,
stress-strain state, displacement, cylindrical tank.
Section 2
Informatics, Computer Engineering and Control
T. V. Yermolenko, D. V. Rolik. CLASSIFICATION OF HEART ABNORMALITIES USING DEEP LEARNING.
UDC 004.891.3 Annotation:The article deals with the problem of diagnosing heart diseases by imaging scalograms of phonocardiographic signals using convolutional neural networks.
The information content of wavelet bases by the entropy criterion, as well as the efficiency of using networks of various architectures for the development of diagnostic
cardiological systems, have been studied. The Morlet wavelet has the highest information content among the studied bases; it is advisable to use InceptionV3 as the basic model
for classifying heartbeat anomalies. Keywords: phonocardiographic signals, wavelet filters, entropy criterion, scalogram, convolutional neural networks.
T. V. Yermolenko, I. Ye. Samorodsky. ANALYSIS OF THE EFFICIENCY OF DEEP NEURAL NETWORKS ARCHITECTURES FOR THE CLASSIFICATION OF PRODUCTS IMAGES.
UDC 004.932.75 Annotation:The article analyzes the effectiveness of using various architectures of deep neural networks in the task of classifying products by their images.
To solve the problems of unbalanced training data and the similarity of objects of different classes, it is proposed to use class balancing and the API-Net
architecture. Research is carried out on the RP2K dataset. Keywords: convolutional neural networks, ResNet, InceptionV3, UMAP algorithm, API-Net, Class Balanced Loss algorithm.
I. N. Savenkov, T. V. Yermolenko, A. V. Tsybik. DEVELOPING A VAD-ALGORITHM BASED ON DEEP LEARNING.
UDC 004.89:004.93 Annotation:The article provides an overview of the signs by which the presence of a speech component in the audio signal is determined, as well as the most well-known algorithms
that detect speech. To classify signal frames into "noise"/"speech" classes, a convolutional network architecture is proposed, the input of which receives an image of the spectrogram
of the frame. Training and testing of the network was carried out on a data set with different types of noise effects taken from audio data bodies that are freely available. Keywords: Voice Activity Detector, acoustic noise, spectrum energy, spectrogram, convolutional neural network.
РУС
ENG
