This article is devoted to a comparative analysis of the resilience of ResNet18 and ResNet50 neural networks to adversarial attacks on training sets. The issue of the importance of ensuring the safety of learning sets is considered, taking into account the growing scope of artificial intelligence applications. The process of conducting an adversarial attack is described using the example of an animal recognition task. The results of two experiments are analyzed. The purpose of the first experiment was to identify the dependence of the number of epochs required for the successful execution of an adversarial attack on the training set on the neural network version of the ResNet architecture using the example of ResNet18 and ResNet50. The purpose of the second experiment was to get an answer to the question: how successful are attacks on one neural network using modified images of the second neural network. An analysis of the experimental results showed that ResNet50 is more resistant to competitive attacks, but further improvement is still necessary.

Keywords: artificial intelligence, computer vision, Reset, ResNet18, ResNet50, adversarial attacks, learning set, learning set security, neural networks, comparative analysis

The article is devoted to the application of language neural networks for matching text descriptions of products. The analysis of methods for comparing text descriptions of products is carried out, the advantages and disadvantages of each method are noted. The method for matching text descriptions of products, based on Bert neural networks, is considered. Experiments and tests on data sets of text descriptions of similar goods from different retail chains are carried out. Conclusions about the quality of matching various networks of the Bert architecture are made.

Keywords: neural networks, transformers, comparison of text descriptions, text analysis, Bert

The paper examines the causes of the decline in the quality of electrical energy in small settlements. A linear multiplicative model for forecasting electrical loads has been developed based on a time series of electricity consumption in one of the small settlements. A posteriori verification of the multiplicative model has been performed, and the efficiency of the developed model for medium-term forecasting tasks has been demonstrated.

Keywords: electricity consumption, small settlement, linear regression model, multiplicative model, forecasting electrical loads, quality of electrical energy

This article is devoted to a comparative analysis of methods for extracting knowledge from texts used to build ontologies. Various extraction approaches are reviewed, such as lexical, statistical, machine learning and deep learning methods, as well as ontology-oriented methods. As a result of the study, recommendations are formulated for choosing the most effective methods depending on the specifics of the task and the type of data being processed.

Keywords: ontology, knowledge extraction, text classification, named entities, machine learning, semantic analysis, model

The reliability of electric power systems is largely determined by the insulation condition of electrical equipment. Insulation damage can lead to power losses, reduced service life of lines and devices, and emergency shutdowns, so insulation diagnostics is critical to prevent technological disruptions. However, traditional approaches to insulation monitoring are often labor-intensive and subjective. In this regard, the role of computer vision and deep learning methods, capable of automatically detecting insulation defects and thereby increasing the efficiency and objectivity of monitoring, is increasing. This study considers the application of modern architectures of deep convolutional neural networks for the problem of recognizing insulating elements of electrical equipment. Particular attention is paid to the comparative analysis of several state-of-the-art models. The considered architectures show effective results and provide deep multi-scale analysis of scene features based on convolutional networks. In this paper, the models are used in conjunction with image augmentation algorithms. Data augmentation allows you to artificially expand limited sets of training images through various transformations, which is especially important for a small dataset. The application of these methods is aimed at improving the quality of training data and reducing the risk of overfitting models, as well as overcoming the imbalance of classes in the sample by generating additional fault samples. The proposed approach includes conducting a sequential comparative experiment on a small and limited set of image data from power facilities. A comparison was made of the accuracy and completeness metrics of various neural network architectures when combining various augmentation strategies in order to identify a combination of models and data augmentation methods that provide the highest recognition accuracy. The results of the study will help determine the most effective augmentation models and techniques suitable for real-life operating conditions at power facilities, taking into account complex backgrounds, variable lighting, and different angles of equipment shooting. Identifying such optimal solutions based on deep learning is intended to improve the reliability and efficiency of automated insulation monitoring in the power industry.

Keywords: computer vision, convolutional neural networks, isolation, defect, data augmentation, machine learning, energy, automation of image analysis

The loading history, including duration, stress level, number of cycles and environmental influences, also affects concrete performance by increasing strength while reducing deformability. Under long-term service, concrete can exhibit elastic behaviour until stresses in the order of 70-80% of the expected compressive strength are reached. The plastic and viscous properties of concrete play an important role under dynamic loading, determining its dynamic hardening capacity. Thus, long-term operation significantly alters the dissipative properties of concrete and affects its response under accidental dynamic effects. The aim of this study was to assess the extent to which the deformed state of the frame as a result of the erection sequence, creep and shrinkage strain accumulation influences the collapse resistance in an emergency situation.The process of deformation of reinforced concrete frame of a multi-storey building in an emergency design situation was modelled, taking into account the stage of construction and different age of concrete at the moment of load application. The computational analysis was performed in quasi-static formulation in Scad Office 21.1.9.9 using the ‘Assembly’ module.According to the results of the study, deformations and forces in the elements of the load-bearing system after the initial local collapse in it have been obtained and analysed. It is shown that when taking into account the sequence of building erection, accumulation of creep and shrinkage deformations, the building resistance to progressive collapse decreases.

Keywords: monolithic reinforced concrete frame, progressive collapse, creep, shrinkage, modulus of elasticity, modulus of deformation, static-dynamic loading

To ensure the safety of capital construction facilities, it is necessary to anticipate and predict risks at the planning and design stage. Risk forecasting is carried out in both quantitative and qualitative measurement. The accuracy of the calculations requires consideration of a large number of different risks, their causes, possible consequences and the likelihood of their occurrence. At this scale of input, traditional ways of calculating construction risks are costly in terms of money and labour, and can be very time-consuming. Artificial intelligence and machine learning technologies automate the process of risk assessment and calculation. With digital technology, all the factors arising during construction will be taken into account in real time. Despite a number of limitations in the application of this technology, this method is the most promising and increasingly widespread.

Keywords: construction, capital engineering, safety, risk, risk forecasting, risk assessment, risk management, artificial intelligence, machine learning

This study presents a comparative analysis of machine learning models used for driver classification based on microelectromechanical system (MEMS) sensor data. The research utilizes the “UAH-DriveSet” open dataset, which includes over 500 minutes of driving data with annotations for aggressive driving events, such as sudden braking, sharp turns, and rapid acceleration. The models evaluated in this study include gradient boosting algorithms, a recurrent neural network and a convolutional neural network. Special attention is given to the impact of data segmentation parameters, specifically window size and overlap, on classification performance using the sliding window method. The effectiveness of each model was assessed based on classification metrics such as accuracy, precision, and F1 score. The results show that gradient boosting “LightGBM” outperforms the other models in terms of accuracy and F1 score, while long short-term memory model demonstrates good performance with time-series data but requires larger datasets for better generalization. Convolutional neural network, while effective for identifying short-term patterns, faced difficulties with class imbalances. This research provides valuable insights into selecting the most appropriate machine learning models for driver behavior classification and offers directions for future work in developing intelligent systems using MEMS sensor data.

Keywords: driver behavior analysis, microelectromechanical system sensors, machine learning, aggressive driving, gradient boosting, recurrent neural networks, convolutional neural networks, sliding window, driver classification

The results of fire dynamics simulation based on the FDS software kernel are a large amount of data describing the dynamics of various parameters in the space of the studied object. Solving various research problems based on them may require quite complex processing, which goes beyond the functionality of existing software solutions. The article is devoted to the method of efficiency increasing for numerical fire dynamics simulation results processing by automating the implementation of relevant operations. The article describes the functional model of the developed technology and its main stages. Approbation of the proposed method was carried out using the example of solving the problem of forming initial data arrays in high spatial and time resolution for the subsequent study of enclosing tunnel structures heating in case of fire. Graphs of the gas medium temperature at various points under the roof of the tunnel structure from the coordinate are presented, as well as temperature fields in the vertical section of the investigated structure in the plane passing through the fire focus at different times. Based on the comparative analysis, it was shown that the speed of calculation results automated processing is several orders of magnitude higher compared to methods that use the functionality of existing software solutions designed to view the output of the fire dynamics simulation.

Keywords: fire dynamics simulation, automation, data processing, tunnel structures, mathematical model, FDS

The article discusses current issues related to the design of a smart home wireless local area network based on splitter-repeater modules. Special attention in the study is paid to the modules of wired and wireless hubs and switches. The results of the comparative characteristics of PLC and FBT splitter-repeaters are also presented. Particular emphasis is placed on the network topology and its main components.

Keywords: wireless network, topology, data, transmission, power, traffic, packet, failures, adapter, cable, connection

The article explores the actor model as implemented in the Elixir programming language, which builds upon the principles of the Erlang language. The actor model is an approach to parallel programming where independent entities, called actors, communicate with each other through asynchronous messages. The article details the main concepts of Elixir, such as comparison with a sample, data immutability, types and collections, and mechanisms for working with the actors. Special attention is paid to the practical aspects of creating and managing actors, their interaction and maintenance. This article will be valuable for researchers and developers interested in parallel programming and functional programming languages.

Keywords: actor model, elixir, parallel programming, pattern matching, data immutability, processes, messages, mailbox, state, recursion, asynchrony, distributed systems, functional programming, fault tolerance, scalability

The transition from scheduled maintenance and repair of equipment to maintenance based on its actual technical state requires the use of new methods of data analysis based on machine learning. Modern data collection systems such as robotic unmanned complexes allow generating large volumes of graphic data in various spectra. The increase in data volume leads to the task of automating their processing and analysis to identify defects in high-voltage equipment. This article analyzes the features of using computer vision algorithms for images of high-voltage equipment of power plants and substations in the infrared spectrum and presents a method for their analysis, which can be used to create intelligent decision support systems in the field of technical diagnostics of equipment. The proposed method uses both deterministic algorithms and machine learning. Classical computer vision algorithms are applied for preliminary data processing in order to highlight significant features, and models based on unsupervised machine learning are applied to recognize graphic images of equipment in a feature space optimized for information space. Image segmentation using a spatial clustering algorithm based on the density distribution of values ​​taking into account outliers allows detecting and grouping image fragments with statistically close distributions of line orientations. Such fragments characterize certain structural elements of the equipment. The article describes an algorithm that implements the proposed method using the example of solving the problem of detecting defects in current transformers, and presents a visualization of its intermediate steps.

Keywords: diversification of management, production diversification, financial and economic purposes of a diversification, technological purposes of ensuring flexibility of production

The annual growth of the load on data centers increases many times over, which is due to the growing growth of users of the information and telecommunications network Internet. Users access various resources and sources, using search engines and services for this. Installing equipment that processes telecommunications traffic faster requires significant financial costs, and can also significantly increase the downtime of the data center due to possible problems during routine maintenance. It is more expedient to focus resources on improving the software, rather than the hardware of the equipment. The article provides an algorithm that can reduce the load on telecommunications equipment by searching for information within a specific subject area, as well as by using the features of natural language and the process of forming words, sentences and texts in it. It is proposed to analyze the request based on the formation of a prefix tree and clustering, as well as by calculating the probability of the occurrence of the desired word based on the three sigma rule and Zipf's Law.

Keywords: Three Sigma Rule, Zipf's Law, Clusters, Language Analysis, Morphemes, Prefix Tree, Probability Distribution

Two capacitive methods of measuring the linear density of one-dimensional fibrous products are considered. The sensitivity of the measurement results to variations in the geometric and physical parameters of the measuring device for the differential and resonance measurement methods is estimated. A weak, almost linear dependence of the measurement error on parameter variations in a wide variation range is established. The good suitability of both methods for measuring the linear density of one-dimensional products by the capacitive method and the high correlation between the measured value and the measurement results are substantiated.

Keywords: fibrous materials, one-dimensional products, linear density, capacitive measurement method, capacitive method, differential circuit, resonant measurement circuit, parameter variations

The article presents the existing methods of reducing the dimensionality of data for teaching machine models of natural language. The concepts of text vectorization and word-form embedding are introduced. The task of text classification is being formed. The stages of classifier training are being formed. A classifying neural network is being designed. A series of experiments is being conducted to determine the effect of reducing the dimension of word-form embeddings on the quality of text classification. The results of evaluating the work of trained classifiers are compared.

Keywords: natural language processing, vectorization, word-form embedding, text classification, data dimensionality reduction, classifier

The paper discusses the use of the M/M/n mass service model to analyze the performance of cloud storage systems. Simulations are performed to identify the impact of system parameters on average latency, blocking probability, and throughput. The results demonstrate how optimizing the number of servers and service intensity can improve system performance and minimize latency. The relevance of the study is due to the need to improve the performance of cloud solutions in the context of growing data volumes and increasing load on storage systems.

Keywords: cloud storage, mass service theory, M/M/n model, Python, modeling, performance analysis

For the development of automated systems for designing ultraviolet irradiators intended to compensate for the deficiency of natural ultraviolet, it is critically important to know the spatial distribution of the erythemal radiation power. However, there are no suitable sensors for direct measurement of this value on the Russian market. In this regard, an alternative method for determining the erythemal radiation power is considered, which does not require the use of specialized erythemal-sensitive receivers. The method is based on obtaining the spatial distribution of the erythemal radiation power by taking into account the curve of the relative spectral erythemal efficiency of radiation and preliminary measurements on a gonioradiometric setup of the distribution of energy illuminance in the UVA (320 - 400 nm), UVB (280 - 320 nm) regions and the relative spectral distribution of the irradiator or radiation source for one arbitrarily selected direction in the wavelength range of 280 - 400 nm.

Keywords: ultraviolet radiation; erythemal radiation; irradiation units; measurement method, radiation strength; spatial distribution of erythemal radiation strength, method

During the research, a new prestressing system was developed for carbon fiber reinforced polymer plates to reinforce damaged steel beams. A parametric analysis was performed using finite element modeling. The results showed that satisfactory amplification efficiency can be achieved using the new pre-voltage system. The prestressed carbon fiber significantly increased the performance when bending beams at the elastic and elastic-plastic stages due to the use of high-strength carbon fiber plates. In addition, as the pre-voltage level increased, the amplification efficiency increased. A simple increase in the area or modulus of elasticity of the carbon fiber plate slightly improved the hardening efficiency, while the simultaneous application of prestressing clearly increases the hardening efficiency.

Keywords: reinforcement, steel beam, prestressing, new system, carbon fiber plate

This paper considered the problem of detection and classification of surface objects in low visibility conditions such as rain and fog. The focus is on the application of state-of-the-art deep learning algorithms, in particular the YOLO architecture , to improve detection accuracy and speed. The introduction to the problem includes a discussion of the limitations of visibility degradation, the change in shape and size of objects depending on the viewing angle, and the lack of training data. The paper also presents the use of discrete wavelet transform to improve image quality and increase the robustness of the systems to adverse conditions. Experimental results show that the proposed algorithm achieves high accuracy and speed, which makes it suitable for application in drone video monitoring systems.

Keywords: YOLO, wavelet transform, overwater objects, drones, low visibility condition, Fourier transforms, Haar

In the work describes the extreme filtering method and the author's approaches that allow adapting it to work in real time: frame-by-frame processing and the method with signal loading. Further, solutions are presented that can be used to implement the above on real devices. The first solution is to use the Multiprocessing library for the Python language. The second approach involves creating a client-server application and sending asynchronous POST requests to implement the frame-by-frame signal processing method. The third method is also associated with the development of a client-server application, but with the WebSocket protocol, not HTTP, as in the previous approach. Then, the results are presented, and conclusions are made about the suitability of the author's approaches and solutions for working on real devices. It is noted that the solution based on the use of the WebSocket protocol is of particular interest. This solution is suitable for both the frame-by-frame signal processing method and the method with value loading. It is also noted that all approaches proposed by the author are workable, which is confirmed by the time values ​​and the coincidence of the graphs.

Keywords: extreme filtering, frame-by-frame signal processing method, method with value loading, Multiprocessing, HTTP, WebSocket, REST, JSON, Python, microcontrollers, single-board computers

The main maintenance of a diversification of production as activity of subjects of managing is considered. being shown in purchase of the operating enterprises, the organizations of the new enterprises, redistribution of investments in interests of the organization and development of new production on available floor spaces. The most important organizational economic targets of a diversification of management are presented by innovative activity of the industrial enterprise.

Keywords: software systems, visualization, data, graphic systems, parts, models, diagrams, drawings

The article is devoted to the developed code designer for the Scilab environment, which is intended to automate the process of creating software modules. The program allows you to generate code for Scilab through an intuitive interface, providing users with tools for working with variables, loops, graphs, system analysis and user-defined functions. The constructor allows you to write programs for Scilab without knowledge of a programming language.

Keywords: Scilab, code designer, programming automation, code generation, visual programming

Linear feedback shift registers (LFSR) and the pseudo-random sequences of maximum length (m-sequences) generated by them have become widely used in solving problems of mathematical modeling, cryptography, radar and communications. The wide distribution is due to their special properties, such as correlation. An interesting, but rarely discussed in the scientific literature of recent years, property of these sequences is the possibility of forming quasi-orthogonal matrices on their basis.In this paper, was conducted a study of methods for generating quasi-orthogonal matrices based on pseudo-random sequences of maximum length (m-sequences). An analysis of the existing method based on the cyclic shift of the m-sequence and the addition of a border to the resulting cyclic matrix is carried out. Proposed an alternative method based on the relationship between pseudo-random sequences of maximum length and quasi-orthogonal Mersenne and Hadamard matrices, which allows generating cyclic quasi-orthogonal matrices of symmetric structure without a border. A comparative analysis of the correlation properties of the matrices obtained by both methods and the original m-sequences is performed. It is shown that the proposed method inherits the correlation properties of m-sequences, provides more efficient storage, and is potentially better suited for privacy problems.

Keywords: orthogonal matrices, quasi-orthogonal matrices, Hadamard matrices, m-sequences

The article considers the options for visual programming of information support means for software and information complexes for UAV operators training. The main criterion indicators for systematically organizing the set of components for reusing program code are identified. An example of an unmanned payload carrier in various representative forms of visualization is given. A comparison of the labor intensity of developing the specified software and information implementations for the same unmanned robotics object with their normative labor intensity is shown. The variants of content filling during the development of the same material part of the considered device for various aspects of training specialists in the management and operation of UAV are considered. The principle of systematization of components by means of ordering the complexity of presentation and softwarе implementation is shown.

Keywords: risk forecasting, information support, training of unmanned aircraft systems operators, labor intensity assessment

The paper presents a method for quantitative assessment of zigzag trajectories of vehicles, which allows to identify potentially dangerous behavior of drivers. The algorithm analyzes changes in direction between trajectory segments and includes data preprocessing steps: merging of closely spaced points and trajectory simplification using a modified Ramer-Douglas-Pecker algorithm. Experiments on a balanced data set (20 trajectories) confirmed the effectiveness of the method: accuracy - 0.8, completeness - 1.0, F1-measure - 0.833. The developed approach can be applied in traffic monitoring, accident prevention and hazardous driving detection systems. Further research is aimed at improving the accuracy and adapting the method to real-world conditions.

Keywords: trajectory, trajectory analysis, zigzag, trajectory simplification, Ramer-Douglas-Pecker algorithm, yolo, object detection