Search

RELEVANCE of this study lies in the use of an artificial neural network to predict the volume of water in the Coca reservoir (Coca hydroelectric power station in Ethiopia). As you know, hydropower, being renewable energy, is one of the technologies that produce electricity with the least impact on global climate change. During this time, Ethiopia received about 87% (4,674 MW) of electricity from hydropower. It is one of the countries affected by the problems of climatic phenomena, such as floods, droughts and hurricanes, which affect the potential of hydropower. THE PURPOSE. In order to maintain safe operation, good production efficiency, better water resources management, effective decision-making, accident prevention and early warning and restrictions on electricity production, water volume forecasting is necessary. Which, in turn, is a nonlinear problem, and a multilinear perceptron-type neural network (MLP) is suitable for this purpose. METHODS. In this study, different models with different selected number of nodes and layers were identified, since there is no specific rule for determining the architecture of an artificial neural network. Statistical analysis (mean square error (MSE) and R-squared (R2)) was used to verify the validity of the model by comparing the actual values of water inflow with the predicted values. results. The inflow prediction was carried out using the ANN method based on a multilayer perceptron (MLP). The performance of each model was evaluated using the mean square error (MSE) and efficiency coefficient (R2), which are among the most commonly used statistical methods in hydrological modeling. CONCLUSION. The results obtained show that the models successfully predicted flood runoff over the reservoir.

The article describes the problems of developing neurocontrollers for controlling dynamic objects, including the complexity of forming training data sets. It is indicated that one of the known methods for training an artificial neural network controlling an object is the neuroevolutionary approach, which involves using a genetic algorithm to adjust the synaptic weighting coefficients of an artificial neural network. The idea of using a means of demonstrating the evolutionary approach to adjusting the weighting coefficients of an artificial neural network for practical training of students in the basics of the neuroevolutionary approach is proposed. Software has been developed to demonstrate the neuroevolutionary approach using the example of the evolution of an artificial neural network of a given structure intended to control a simplified computer model of an autonomous vehicle. A method for resolving the problem of stagnation when using the evolutionary approach to training an artificial neural network is described. Options for using the developed software in teaching students the basics of artificial intelligence technologies and evolutionary methods of multicriteria optimization are proposed.

Artificial intelligence has become widespread in image processing tasks. At the same time, the number of vulnerabilities is increasing in systems implementing these artificial intelligence technologies (the attack surface is increasing). The main threats to information security can be implemented by introducing malicious perturbations into the input data, regardless of their type. To detect such attacks, approaches and methods have been developed based, in particular, on the use of an auto-encoder or the analysis of layers of the target neural network. The disadvantage of existing methods, which significantly reduce the scope of their application, is binding to the dataset or model architecture. This paper discusses the issues of expanding the scope (increasing scalability) of methods for detecting L0-optimized perturbations introduced by unconventional pixel attacks. An approach to detecting these attacks using statistical analysis of input data, regardless of the model and dataset, is proposed. It is assumed that the pixels of the perturbation embedded in the image, as a result of the L0-optimized attack, will be considered both local and global outliers. Outlier detection is performed using statistical metrics such as deviation from nearest neighbors and Mahalanobis distance. The evaluation of each pixel (anomaly score) is performed as a product of the specified metrics. A threshold clipping algorithm is used to detect an attack. When a pixel is detected for which the received score exceeds a certain threshold, the image is recognized as distorted. The approach was tested on the CIFAR-10 and MNIST datasets. The developed method has demonstrated high accuracy in detecting attacks. On the CIFAR-10 dataset, the accuracy of detecting onepixel attack (accuracy) was 94.3 %, and when detecting a Jacobian based Saliency Map Attack (JSMA) — 98.3 %. The proposed approach is also applicable in the detection of modified pixels. The proposed approach is applicable for detecting one-pixel attacks and JSMA, but can potentially be used for any L0-optimized distortions. The approach is applicable for color and grayscale images regardless of the dataset. The proposed approach is potentially universal for the architecture of a neural network, since it uses only input data to detect attacks. The approach can be used to detect images modified by unconventional adversarial attacks in the training sample before the model is formed.

Using of existing methods of determining the characteristics of soils which are part of current regulatory documents and which are based on the hypothesis of normal character of distribution require considerable time and material costs. According to the results of conducted laboratory researches the hypothesis wasn’t confirmed. In the paper it proposes to use trained artificial neural network for determination of the deformation modulus of binary composite «sand — granules of expanded polystyrene». Thus, it has been proven efficiency proposing method using trained artificial neural network in compare classical regression equation for determination of the deformation modulus of the binary composite. With a confidence probability of P = 95 % the absolute value of the relative error is equal to 11,8 % the proposing learning artificial neural network in 11 times less than the absolute value of the relative error of classical regression equation. Also with a confidence probability of P = 95 % the coefficient of determination is equal to 0,5641 and in 6,6 times less than it of regression equation. Further research will be directed to the selection of the values of the parameters of the artificial neural network program for increase the accuracy of determining the deformation modulus of the binary composite.

Introduction. This work presents a methodology for detecting inter-turn short circuit (ITSC) and broken rotor bars (BRB) fault in variable speed induction machine controlled by field oriented control. If any of these faults are not detected at an early stage, it may cause an unexpected shutdown of the industrial processes and significant financial losses. Purpose. For these reasons, it is important to develop a new diagnostic system to detect in a precautionary way the ITSC and BRB at various load condition. We propose the application of discrete wavelet transform to overcome the limitation of traditional technique for no-stationary signals. The novelty of the work consists in developing a diagnosis system that combines the advantages of both the discrete wavelet transform (DWT) and artificial neural network (ANN) to identify and diagnose defects, related to both ITSC and BRB faults. Methods. The suggested method involves analyzing the electromagnetic torque signal using DWT to calculate the stored energy at each level of decomposition. Then, this energy is applied to train neural network classifier. The accuracy of ANN based on DWT, was improved by testing different orthogonal wavelet functions on simulated signal. The selection process identified 5 pertinent wavelet energies, concluding that, Daubechies44 (db44) is the best suitable mother wavelet function for effectively detecting and classifying failures in machines. Results. We applied numerical simulations by MATLAB/Simulink software to demonstrate the validity of the suggested techniques in a closed loop induction motor drive. The obtained results prove that this method can identify and classify these types of faults under various loads of the machine. References 31, table 1, figures 9.

Information security risk assessment is a crucial component of industrial management techniques that aids in identifying, quantifying, and evaluating risks in comparison to criteria for risk acceptance and organizationally pertinent objectives. Due to its capacity to combine several parameters to determine an overall risk, the traditional fuzzy-rule-based risk assessment technique has been used in numerous industries. The technique has a drawback because it is used in situations where there are several parameters that need to be evaluated, and each parameter is expressed by a different set of linguistic phrases. In this paper, fuzzy set theory and an artificial neural network (ANN) risk prediction model that can solve the issue at hand are provided. Also developed is an algorithm that may change the risk-related factors and the overall risk level from a fuzzy property to a crisp-valued attribute is developed. The system was trained by using twelve samples representing 70%, 15%, and 15% of the dataset for training, testing, and validation, respectively. In addition, a stepwise regression model has also been designed, and its results are compared with the results of ANN. In terms of overall efficiency, the ANN model (R2= 0.99981, RMSE=0.00288, and MSE=0.00001,) performed better, though both models are satisfactory enough. It is concluded that a risk-predicting ANN model can produce accurate results as long as the training data accounts for all conceivable conditions.

It is widely known that among sound signals generated by natural and anthropogenic phenomena, the most long-lived are waves of frequency less than 20 Hz, called infrasound. This property allows tracking at a distance by infrasound monitoring the occurrence of high-energy events on regional scales (up to 200–300 km). At the same time, the separation of useful infrasound signals from background noise is a non-trivial task in real-time and post-facto signal processing. In this paper we propose a new method for classification of specific signals in infrasound monitoring data using Shannon permutation entropy and vectors of frequency distribution of occurrence frequencies of permutations of consecutive sample values of rank 3 (number of permutation elements). To evaluate the validity of the proposed entropy-based classification method, two machine learning methods — random forest method and classical neural network approach — implemented in Python language using Scikit-lean, TensorFlow and Keras libraries were used. The classification quality was evaluated against the traditional frequency-based method of class extraction based on Fourier transform. Recognition was performed on the prepared infrasound monitoring data in the Altai Republic. The results of computational experiment on the separation of 5 classes of signals showed that classification by the proposed method gives the same results of recognition by neural network with in comparison with frequency classification of the original data; the recognition accuracy was 51–58 %. For the random forests method, the recognition accuracy of frequency classes was slightly higher: 51 % vs. 45 % for classes using the permutation entropy method. The analysis of the results of the computational experiment shows sufficient competitiveness of the method of classification by permutation entropy in the recognition of infrasound signals. In addition, the proposed method is much easier to implement for inline signal processing in lowconsumption microcontroller systems. The next step is to test the method at infrasound signal registration points and as part of the infrasound monitoring data processing system for real-time event detection.

Objectives. A valley is a region of an objective function landscape in which the function varies along one direction more slowly than along other directions. In order to determine the error of the objective function minimum location in such regions, it is necessary to analyze relations of valley parameters.Methods. A special test function was used in numerical experiments to model valleys with variables across wide ranges of parameters. The position and other valley parameters were defined randomly. Valley dimensionality and ratio were estimated from eigenvalues of the approximated Hessian of objective function in the termination point of minimum search. The error was defined as the Euclidian distance between the known minimum position and the minimum search termination point. Linear regression analysis and approximation with an artificial neural network model were used for statistical processing of experimental data.Results. A linear relation of logarithm of valley ratio to logarithm of minimum position error was obtained. Here, the determination coefficient R2 was ~0.88. By additionally taking into account the Euclidian norm of the objective function gradient in the termination point, R2 can be augmented to ~0.95. However, by using the artificial neural network model, an approximation R2 ~ 0.97 was achieved.Conclusions. The obtained relations may be used for estimating the expected error of extremum coordinates in optimization problems. The described method can be extended to functions having a valley dimensionality of more than one and to other types of hard-to-optimize algorithms regions of objective function landscapes.

ABSTRACT: Introduction. The industrial activities result in the release of a significant number of pollutants into the atmosphere, which have a negative impact on human health and the environment. The development of sustainable principles for construction sites development requires the elaboration of algorithms, assessment methodologies, and monitoring of the environmental situation at the construction site. Analysis of research works in the field of environmental monitoring of air pollution by hazardous substances has shown that the speed and reliability of the obtained information are crucial in polluted atmospheric conditions. In this article, we examine the application of an intelligent environmental monitoring system with Salavat city Republic of Bashkortostan as the example. The article presents an example of continuous eco-monitoring using three automatic air pollution monitoring stations for controlling the atmospheric air in residential areas. Methods and Materials. An intelligent eco-monitoring system is proposed, which is capable of assessing the state of the air basin of an industrial city, identifying sources of increased air pollution, processing information using an artificial neural network in real-time, generating control actions on sources of increased air pollution, and providing recommendations to businesses for optimizing their operations to prevent exceeding the maximum allowable concentrations of harmful substances. Results. The analysis of air quality in Salavat has led to the conclusion that it is necessary to install window filter-screens with nanofiber-based filtering elements in residential buildings. Discussion. The results of the air basin monitoring can be used in an integrated intelligent environmental monitoring system for assessing environmental conditions and can be made available to interested parties. Considering the results in Salavat, the project can be replicated for use in eco-monitoring systems of towns and cities in the Russian Federation. Conclusion. As a result of the project, the program utilizing artificial intelligence has been developed, which will contribute to reducing the level of air pollution in the future.

ABSTRACT: Introduction. Approximately 20% of accidents in buildings and structures are associated with the destruction of metal structures. Nanocoatings applied to metal structures significantly improve their operational properties, but at the same time make it difficult to use traditional non-destructive testing methods that require access to the surface of the base metal. The quality of the nanocoating is controlled during its formation. To prevent the destruction of metal structures, it is necessary to identify structural heterogeneities of the base metal in micro- and sub-micron sizes, which arise in areas of increased mechanical stress concentration and transform into macrodefects leading to structural failure. Methods and tools are required to non-destructively identify the stress-strain state and damage of the base metal structure through the layer of nanocoating. Electromagnetic-acoustical (EMA) transformation, is considered promising for solving this problem. Existing EMA tools do not fully utilize the potential of EMA conversion, have insufficient sensitivity and the lack of informativity. Methods and materials. For experimental research, we selected widely used in construction structures steels. Research was conducted on the interrelationship of structural alterations in standard samples of metals and the parameters of the electro-magnetic (EMA) transducer under static and cyclic loads. Results. Load diagrams of metal samples were obtained, with accompanying photographs of microstructure in control points, along with their frequency models derived through spectral analysis of the signal from EMA transducer. Discussion. Changes in the stress-strain state of the metal and accumulation of damage in its structure result in a complex interrelated set of changes in mechanical, acoustic, and electro-physical properties, all of which are reflected in the changes in the parameters of the frequency model. Conclusion. Based on the results of the conducted research, the use of a frequency model has been proposed as an integral parameter for identifying the stress-strain state and damage of metal in the equipment. The use of an artificial neural network for analyzing frequency model parameters simplifies the process of identifying the stress-strain state and damage of metal structures and increases its reliability.

The article describes an approach to implementing the ability of intelligent systems to self-learn through an intelligent framework for the development of artificial neural networks. A method of interaction between intelligent systems and a framework based on a library of reusable components is proposed, which allows to automatically design and train artificial neural networks based on a given problem specification. Based on the analysis of the activity of developers of artificial neural networks, a model of the problem solver of such a framework is described in the form of a hierarchy of actions for the development of artificial neural networks.

Introduction. At present, an intuitive graphical interface is an indispensable component of modern agricultural-oriented software products. Aim of the Article. The research is aimed at improving the efficiency of sorting apples by developing a graphical control interface for a vision system to recognize various defects and sort apples. Materials and Methods. The authors used methods of analysis, enumeration, comparison and synthesis of modern software solutions. Results. As a result of the research, there was created a graphical application of the software for the control unit of the machine vision system containing all the necessary tools for managing and optimizing costs when sorting apples into three commercial quality classes. Discussion and Conclusion. The graphical interface of the machine vision system was used in the line LSP-4 for sorting and packing apples. It was developed by Scientific and Practical Center NAS of Belarus for Agricultural Mechanization in 2020 and successfully passed state acceptance tests.

Systems implementing artificial intelligence technologies have become widespread due to their effectiveness in solving various applied tasks including computer vision. Image processing through neural networks is also used in securitycritical systems. At the same time, the use of artificial intelligence is associated with characteristic threats including disruption of machine learning models. The phenomenon of triggering an incorrect neural network response by introducing perturbations that are visually imperceptible to a person was first described and attracted the attention of researchers in 2013. Methods of attacks on neural networks based on malicious perturbations have been continuously improved, ways of disrupting the operation of neural networks in processing various types of data and tasks of the target model have been proposed. The threat of disrupting the functioning of neural networks through these attacks has become a significant problem for systems implementing artificial intelligence technologies. Thus, research in the field of countering attacks based on malicious perturbations is very relevant. This article describes current attacks, provides an overview and comparative analysis of such attacks on image processing systems based on artificial intelligence. Approaches to the classification of attacks based on malicious perturbations are formulated. Defense methods against such attacks are considered, their shortcomings are revealed. The limitations of the applied defense methods that reduce the effectiveness of counteraction to attacks are shown. Approaches and practical measures to detect and eliminate harmful disturbances are proposed.

The paper presents the results of studies of the predictive models development based on retrospective data on planned electricity consumption in the region with a significant share of enterprises in the mineral resource complex. Since the energy intensity of the industry remains quite high, the task of rationalizing the consumption of electricity is relevant. One of the ways to improve control accuracy when planning energy costs is to forecast electrical loads. Despite the large number of scientific papers on the topic of electricity consumption forecasting, this problem remains relevant due to the changing requirements of the wholesale electricity and power market to the accuracy of forecasts. Therefore, the purpose of this study is to support management decisions in the process of planning the volume of electricity consumption. To realize this, it is necessary to create a predictive model and determine the prospective power consumption of the power system. For this purpose, the collection and analysis of initial data, their preprocessing, selection of features, creation of models, and their optimization were carried out. The created models are based on historical data on planned power consumption, power system performance (frequency), as well as meteorological data. The research methods were: ensemble methods of machine learning (random forest, gradient boosting algorithms, such as XGBoost and CatBoost) and a long short-term memory recurrent neural network model (LSTM). The models obtained as a result of the conducted studies allow creating short-term forecasts of power consumption with a fairly high precision (for a period from one day to a week). The use of models based on gradient boosting algorithms and neural network models made it possible to obtain a forecast with an error of less than 1 %, which makes it possible to recommend the models described in the paper for use in forecasting the planned electricity power consumption of united power systems.

Relevance. Achieving the greatest rate of penetration is the aim of every drilling engineer because it is one of the most significant factors influencing drilling costs. However, a variety of drilling conditions could have an impact on rate of penetration, complicating its forecast. Aim. To suggest a novel strategy to accurately predict rate of penetration and optimize drilling parameters. Objects. Real-time drilling data of a few wells in the Ca Tam oil field, Vietnam, with more than 900 datasets including significant parameters like rotary speed, weight on bit, standpipe pressure, flow rate, weight of mud, torque. Methods. Various methods using Artificial Neural Network was proposed to estimate rate of penetration. Results. The number of neurons in a hidden layer was varied then the results of different Artificial Neural Network models were compared in order to obtain the optimal model. The final Artificial Neural Network model shows high exactness when contrasted with actual rate of penetration, in this manner, it tends to be suggested as a successful and reasonable approach to predict the rate of penetration of different wells in the Ca Tam oil field. Based on the proposed Artificial Neural Network model, the optimal weight on bit was determined for the drilling interval from 1800 to 2300 m of oil wells in the research region.

Background. To enhance efficiency in obtaining plant biomass via the research into existing techniques and development of novel ones is a challenging task. The purpose of the work is to explore increasing productivity and efficiency of nutrient absorption by means of artificial intelligence algorithms and biosensors in the information-measuring and aeroponic cultivation control system for growing plants in urban areas. Materials and methods. Artificial neural network algorithms along with specialized sensors inserted in the root zone of control plants have been proposed. Results. The review analysis results for the existing methods of growing plants have been presented. Major advantages and disadvantages of these methods in terms of specific parameters have been considered. The most effective method has been identified, and novel techniques to increase the efficiency in obtaining plant biomass have been proposed. Artificial intelligence architecture for the information-measuring and aeroponic cultivation control system for growing plants in urban areas has been developed. Conclusions. It has been revealed that provided the proposed techniques were implemented, it would advance the aeroponic method for growing plants in urban areas. The use of artificial intelligence algorithms and specialized sensors in the information-measuring and aeroponic cultivation control system for growing plants would ensure increased productivity and efficiency of nutrient absorption, and reduce maintenance requirements.

The article analyzes the possibility of using a Siamese convolutional neural network to solve the problem of vascular authentication on an embedded hardware platform with limited computing resources (Orange Pi One). The authors give a brief review of modern methods for calculating image feature vectors used in the tasks of classifying, comparing or searching for images by content: based on variational series (histograms), local descriptors, singular point descriptors, descriptors based on hash functions, neural network descriptors. They suggest using the architecture of a biometric authentication system (BAS) based on images of palms in the visible and near-IR spectra based on a Siamese convolutional neural network. The developed software solution allows using the Siamese neural network in the "full network" (both symmetrical channels of the neural network are used) and "half of the neural network" (only one channel is used) modes to reduce the time for comparing biometric data vectors - images of the palms of registered BAS users. The authors demonstrate advantages of the neural network features: universality, scalability and competitiveness, including on embedded hardware and software solutions with limited computing resources without graphics accelerators. The studies have shown that using the Siamese neural network, the "overall accuracy" of palm image classification can be improved from 0.929 to 0.968 when compared with the image vectorization method based on a perceptual hash, while showing a comparable authentication time for individuals registered in BAS. In the experiments, the authors use a database of 2,000 images for 400 people.

Background. This article presents the composition and algorithm of the automated control system for the parameters of a pneumatic arch-breaking system installed on bunkers with bulk materials. The relevance and practical significance of the issue considered in the article is determined by the fact that when unloading bulk material from the hopper, various violations of the technological process may occur, manifested in the formation of material freezes inside the container, a decrease in the amount of unloaded material, segregation, up to a complete stop of unloading due to the formation of arches. Materials and/or methods. Methods of mathematical modeling, system analysis, comparison, systems theory, as well as architecture and mathematical model of a recurrent neural network with feedback are used. Results. The composition of input and output parameters of the control system is substantiated. Methods for solving problems of classifying parameter sets are considered and an artificial recurrent neural network with feedback and with a sigmoidal activation function of neurons of the hidden layer and a linear activation function of the output layer is selected. Conclusion. The article analyzes the requirements for the system of automated control of the parameters of the pneumatic arch-breaking system. Based on the analysis of possible states of the system, the composition of the system elements and the sets of input, intermediate and output parameters of the system, as well as the objective function of the system operation, are substantiated. It is shown that the system can be built on the basis of an artificial neural network. The algorithm of the automated control system for the parameters of the pneumatic arch-breaking system has been developed.

Objective. Improving the quality of detection and stego-detection of latent images embedded in the protected object of intellectual property by various methods. Method. The method for stego-detection of latent images based on deep learning is proposed. The method is based on the use of the VGG16 convolutional neural network model, in which the architecture and training parameters are optimized. Result. Increasing the accuracy of detecting stegocontainer images by 3.8%, as well as the possibility of using the algorithm of the developed method for images with a higher resolution than the dimension of the input of an artificial neural network. Conclusion. The developed method is intended for stegan detection in two cases: to detect the fact of illegal use of intellectual property objects; for use in computer forensics when identifying images containing hidden and prohibited information.

Introduction. The widespread use of photovoltaic systems in various applications has spotlighted the pressing requirement for reliability, efficiency and continuity of service. The main impediment to a more effective implementation has been the reliability of the power converters. Indeed, the presence of faults in power converters that can cause malfunctions in the photovoltaic system, which can reduce its performance. Novelty. This paper presents a technique for diagnosing open circuit failures in the switches (IGBTs) of power converters (DC-DC converters and three-phase inverters) in a grid-connected photovoltaic system. Purpose. To ensure supply continuity, a fault-diagnosis process is required throughout all phases of energy production, transfer, and conversion. Methods. The diagnostic approach is based on artificial neural networks and the extraction of features corresponding to the open circuit fault of the IGBT switch. This approach is based on the Clarke transformation of the three-phase currents of the inverter output as well as the calculation of the average value of these currents to determine the exact angle of the open circuit fault. Results. This method is able to effectively identify and localize single or multiple open circuit faults of the DC-DC converter IGBT switch or the three-phase inverter IGBT switches.

Introduction. Cascaded H-bridge multilevel inverters (CHB-MLI) are becoming increasingly used in applications such as distribution systems, electrical traction systems, high voltage direct conversion systems, and many others. Despite the fact that multilevel inverters contain a large number of control switches, detecting a malfunction takes a significant amount of time. In the fault switch configurations diode included for freewheeling operation during open-fault condition. During short circuit fault conditions are carried out by the fuse, which can reveal the freewheeling current direction. The fault category can be identified independently and also failure of power switches harmed by the functioning and reliability of CHB-MLI. This paper investigates the effects and performance of open and short switching faults of multilevel inverters. Output voltage characteristics of 5 level MLI are frequently determined from distinctive switch faults with modulation index value of 0.85 is used during simulation analysis. In the simulation experiment for the modulation index value of 0.85, one second open and short circuit faults are created for the place of faulty switch. Fault is identified automatically by means of artificial neural network (ANN) technique using sinusoidal pulse width modulation based on distorted total harmonic distortion (THD) and managed by its own. The novelty of the proposed work consists of a fast Fourier transform (FFT) and ANN to identify faulty switch. Purpose. The proposed architecture is to identify faulty switch during open and short failures, which has to be reduced THD and make the system in reliable operation. Methods. The proposed topology is to be design and evaluate using MATLAB/Simulink platform. Results. Using the FFT and ANN approaches, the normal and faulty conditions of the MLI are explored, and the faulty switch is detected based on voltage changing patterns in the output. Practical value. The proposed topology has been very supportive for implementing non-conventional energy sources based multilevel inverter, which is connected to large demand in grid.

These days, the ability to predict the result of the development of the system is the guarantee of the successful functioning of the system. Improving the quality and volume of information, complicating its presentation, the need to detect hidden connections makes it ineffective, and most often impossible, to use classical statistical forecasting methods. Among the various forecasting methods, methods based on the use of artificial neural networks occupy a special place. The main objective of our work is to create a neural network that predicts the risk of depression in a person using data obtained using a motor control performance testing system. The stop-signal paradigm (SSP) is an experimental technique to assess a person’s ability to activate deliberate movements or inhibit movements that have become inadequate to external conditions. In modern medicine, the SSP is most commonly used to diagnose movement disorders such as Parkinson’s disease or the effects of stroke. We hypothesized that SSP could serve as a basis for detecting the risk of affective diseases, including depression. The neural network we are developing is supposed to combine such behavioral indicators as: the amount of missed responses, amount of correct responses, average time, the amount of correct inhibition of movements after stopsignal onset. Such a combination of indicators will provide increased accuracy in predicting the presence of depression in a person. The artificial neural network implemented in the work allows diagnosing the risk of depression on the basis of the data obtained in the stop-signal task. An architecture was developed and a system was implemented for testing motor control indicators in humans, then it was tested in real experiments. A comparison of neural network technologies and methods of mathematical statistics was carried out. A neural network was implemented to diagnose the risk of depression using stop-signal paradigm data. The efficiency of the neural network (in terms of accuracy) was demonstrated on data with an expert assessment for the presence of depression and data from the motor control testing system.

This article presents the development and study of a model for formalizing the decision-making process in a computer game using artificial intelligence methods. Game artificial intelligence is implemented using a rather limited set of tools: artificial neural networks, evolutionary algorithms, specialized architectures for building artificial intelligence systems. But neural networks also have a number of disadvantages. Perhaps the most significant of them is the complexity and long network training time. The problem is raised of developing new mathematical algorithms that have a probability of making the right decision, comparable to neural networks, but with less training time. At the moment, there is no single model of the artificial immune system for the development of artificial intelligence in computer and mobile games. Thus, the problem is raised of developing new algorithms that allow realizing interaction with the player in a computer game, having speed and realism comparable to artificial neural networks and expert systems, and at the same time having less training time. One of the ways to solve this problem is to develop a game artificial intelligence model based on an artificial immune system. The dot game is a modification of the Japanese game go. In this game, positional strategic planning is combined with tactical enumeration of options. The implementation of artificial intelligence in this game in the form of a decision tree is not rational for solving this problem. To simplify the search for the optimal solution, heuristic rules were used. To reduce the solution search time, game artificial intelligence was created using methods based on the principles of the immune system. The artificial immune system is an idealized version of the natural analogue and reproduces the key components of the natural process: selection of the best antibodies in the population depending on the degree of their affinity (proximity) to the antigen, antibody cloning, antibody mutation.

Introduction. This research work focuses on the design and experimental validation of fault detection techniques in grid-connected solar photovoltaic system operating under Maximum Power Point Tracking mode and subjected to various operating conditions. Purpose. Six fault scenarios are considered in this study including partial shading, open circuit in the photovoltaic array, complete failure of one of the six IGBTs of the inverter and some parametric faults that may appear in controller of the boost converter. Methods. The fault detection technique developed in this work is based on artificial neural networks and uses discrete wavelet transform to extract the features for the identification of the underlying faults. By applying discrete wavelet transform, the time domain inverter output current is decomposed into different frequency bands, and then the root mean square values at each frequency band are used to train the neural network. Results. The proposed fault diagnosis method has been extensively tested on the above faults scenarios and proved to be very effective and extremely accurate under large variations in the irradiance and temperature. Practical significance. The results obtained in the binary numerical system allow it to be used as a machine code and the simulation results has been validated by MATLAB / Simulink software.

The article deals with the concept of monitoring based on nonparametric statistics. It is supposed the possible application to monitor the technological systems of nuclear power plants. Modern complex plants and technogenic objects are equipped with industrial monitoring systems. The evolution of such systems in natural form is the generation of alternative concepts, constructs, and items selection for efficiency during operation

Introduction. One of the most significant tasks facing road experts is to maintain the transport network in good condition. The process of selecting an appropriate approach to providing such condition is quite complex since it requires considering many parameters, such as the existing condition of the pavement, road category, weather conditions, traffic volume, etc. Recently, the rising trend of digitization in the industry has contributed to the use of artificial intelligence to address problems in several fields, including the bodies in charge of operational control over the status of roadways. Within the context of any control system, the main task of the control system is to carry out reliable forecasting of the operational state of the road in the medium and long term.Materials and Methods. This study investigated the possibility of using artificial neural networks to assess existing pavement characteristics and their potential application in developing road maintenance strategies. A back-propagation neural network was implemented, trained using data from 1,614 investigated sections of the M4 «DON» highway in the road network of the Russian Federation in the period from 2014 to 2018. Several models were developed and trained using the MATLAB application, each with a different number of neurons in the hidden layers.Results. The results of the models showed a convergence between the inferred paving state values and the actual values, as the multiple correlation coefficient (R2) values exceeded 92 % for most of the models during all learning stages.Discussion and Conclusions. The findings suggest that public road authorities may utilize the established models to choose the best road maintenance strategy and assign the most efficient steps to restore road bearing capacity and operation.

Background. Currently accepted risk assessments of periodontitis progression are determinants of indirect stability: periodontal pockets, persistent bleeding of the gums, tooth mobility, local risk factors. In the era of case-oriented medicine, a relevant solution would be to choose periodontal therapy according to one-time consideration of the maximum available range of individual risk factors rather than on general clinical guidelines.Objectives. The study was aimed at determining the relative risk of periodontitis progression after active basic therapy using neural network modeling.Methods. A cohort retrospective study was performed on 109 patients of both sexes, aged 30 to 70 years, after basic treatment of chronic periodontitis (mild, moderate and severe) in the period from 1999 to 2016, who were on supportive periodontal therapy (SPT) for 5 years ≤SPT≤ 20 years. The authors considered data from objective examination of the periodontium and categorical indices (24 in total) assessed before treatment, 4–6 months after basic (active) treatment and 5 years ≤SPT≤ 20 years. Following the analysis of descriptive statistics, target quantitative indices were determined for prognostic modeling of treatment outcomes in periodontitis patients and calculating the residual risk of disease progression. Statistical processing of obtained data was carried out using the Statistica 13.3 package (Tibco, USA). Mean values of the indicators at different time points were compared by means of Wilcoxon’s and Signs criteria; Spearman’s rank correlation coefficient was used to evaluate relevance between predictors and target indicators. The level of statistical significance p = 0.05 was accepted in all cases of analysis. DataMining, an automated neural network of Statistica software, was used as a tool to build neural network models. The task of classifying the level of risk of disease progression was solved by means of ROC analysis. The prognostic potential of the model was assessed using sensitivity and specificity measures.Results. The heterogeneous dynamics of predictor variables describing the state of the periodontium was determined. The outcomes of regenerative periodontal surgery, regardless of gender, age of patients and comorbidities, significantly outperformed those of other approaches, due to the formation of a new dentogingival attachment, although to different extent. Another positive functional outcome was recorded in restoring the dentition integrity by implantation, without any mutually damaging effects. Since revealing the interrelationships between indicators is not equivalent to the predictive value, prognostic models were built for target indicators and stratification of the relative risk of periodontitis progression using automated neural networks. The networks with the best prognostic properties were selected out of 1000 automatically built and trained neural networks — double-layer perceptrons. The sensitivity of the relative risk prognostic model on the training, control and test samples made up 90%, 67%, 80%; the specificity of the model made up 81.481%, 85.714%, 100%. Overall, in the cohort, the sensitivity and specificity accounted for 85.937% and 86.666%. The area under the curve (ROC AUC) is 0.859.Conclusion. The use of an artificial intelligence algorithm for the construction of neural networks for target predictors and stratification of the relative risk of periodontitis progression has advantages over classical methods — it is instrumental in solving classification and regression problems with categorical and quantitative predictor variables using data of arbitrary nature of large and small volumes. The practical implementation of the study results is reflected in the development of a relative risk calculator based on a written computer program.

Introduction. To attain high efficiency and reliability in the field of clean energy conversion, power electronics play a significant role in a wide range of applications. More effort is being made to increase the dependability of power electronics systems. Purpose. In order to avoid any undesirable effects or disturbances that negatively affect the continuity of service in the field of energy production, this research provides a fault detection technique for insulated-gate bipolar transistor open-circuit faults in a three-level diode-clamped inverter of a wind energy conversion system predicated on a doubly-fed induction generator. The novelty of the suggested work ensures the regulation of power exchanged between the system and the grid without faults, advanced intelligence approaches based on a multilayer artificial neural network are used to discover and locate this type of defect; the database is based on the module and phase angle of three-phase stator currents of induction generators. The proposed methods are designed for the detection of one or two open-circuit faults in the power switches of the side converter of a doubly-fed induction generator in a wind energy conversion system. Methods. In the proposed detection method, only the three-phase stator current module and phase angle are used to identify the faulty switch. The primary goal of this fault diagnosis system is to effectively detect and locate failures in one or even more neutral point clamped inverter switches. Practical value. The performance of the controllers is evaluated under different operating conditions of the power system, and the reliability, feasibility, and effectiveness of the proposed fault detection have been verified under various open-switch fault conditions. The diagnostic approach is also robust to transient conditions posed by changes in load and speed. The proposed diagnostic technique's performance and effectiveness are both proven by simulation in the SimPower /Simulink® MATLAB environment.

This paper analyzes the problems of document image recognition and the existing solutions. Document recognition algorithms have been studied for quite a long time, but despite this, currently, the topic is relevant and research continues, as evidenced by a large number of associated publications and reviews. However, most of these works and reviews are devoted to individual recognition tasks. In this review, the entire set of methods, approaches, and algorithms necessary for document recognition is considered. A preliminary systematization allowed us to distinguish groups of methods for extracting information from documents of different types: single-page and multi-page, with text and handwritten contents, with a fixed template and flexible structure, and digitalized via different ways: scanning, photographing, video recording. Here, we consider methods of document recognition and analysis applied to a wide range of tasks: identification and verification of identity, due diligence, machine learning algorithms, questionnaires, and audits. The groups of methods necessary for the recognition of a single page image are examined: the classical computer vision algorithms, i.e., keypoints, local feature descriptors, Fast Hough Transforms, image binarization, and modern neural network models for document boundary detection, document classification, document structure analysis, i.e., text blocks and tables localization, extraction and recognition of the details, post-processing of recognition results. The review provides a description of publicly available experimental data packages for training and testing recognition algorithms. Methods for optimizing the performance of document image analysis and recognition methods are described.

Modern technologies for the development and operation of neural networks are vulnerable to computer attacks with the introduction of software backdoors. Program backdoors can remain hidden indefinitely until activated by input of modified data containing triggers. These backdoors pose a direct threat to the security of information for all components of the artificial intelligence system. Such influences of intruders lead to a deterioration in the quality or complete cessation of the functioning of artificial intelligence systems. This paper proposes an original method for protecting neural networks, the essence of which is to create a database of ranked synthesized backdoor’s triggers of the target class of backdoor attacks. The proposed method for protecting neural networks is implemented through a sequence of protective actions: detecting a backdoor, identifying a trigger, and neutralizing a backdoor. Based on the proposed method, software and algorithmic support for testing neural networks has been developed that allows you to identify and neutralize computer backdoor attacks. Experimental studies have been carried out on various dataset-trained convolutional neural network architectures for objects such as aerial photographs (DOTA), handwritten digits (MNIST), and photographs of human faces (LFW). The decrease in the effectiveness of backdoor attacks (no more than 3 %) and small losses in the quality of the functioning of neural networks (by 8–10 % of the quality of the functioning of a neural network without a backfill) showed the success of the developed method. The use of the developed method for protecting neural networks allows information security specialists to purposefully counteract computer backdoor attacks on artificial intelligence systems and develop automated information protection tools.

This article presents the development and study of a model for formalizing the decision-making process in a computer game using artificial intelligence methods. Currently, various artificial neural networks and expert systems have been created and used for this purpose. The analysis of these works showed that these methods show good results, but they have a number of disadvantages, the most significant of which is the complexity of the organization and the long time spent on training the neural network. Thus, the problem of developing new algorithms is posed that is comparable in decision-making speed with artificial neural networks and expert systems and at the same time has less training time. One way to solve this problem is to develop a decision-making model based on an artificial immune system. The aim of the work is to develop and study a model for formalizing the decision-making process in a computer game using artificial intelligence methods. To demonstrate the possibilities of using the artificial immune system in computer games, the game "dots" was chosen. This game belongs to the category of logical ones, in which only two people can play by default. In this game, positional strategic planning is combined with tactical enumeration of options. To organize the search for a solution in the decision tree, a gaming artificial intelligence was created using methods based on the principles of the immune system. The artificial immune system is an idealized version of the natural analogue and reproduces the key components of the natural process: selection of the best antibodies in the population depending on the degree of their affinity (proximity) to the antigen, antibody cloning, antibody mutation. An artificial immune system represents an idealized version of a natural analogue and reproduces the key components of a natural process: selection of the best antibodies in a population depending on the degree of their affinity (proximity) to an antigen, cloning of antibodies, and mutation of antibodies.

The features of the organization of distance learning of students in a higher educational institution, as well as the information and educational technologies necessary for this, are considered. A system of automatic assessment of students’ knowledge is proposed. It is based on a model in the form of an artificial neural network. The features of such a model are given. The two implemented methods for constructing artificial neural networks have been used in the software module for testing students’ knowledge. The choice of the type of network, its structure, and parameters has been substantiated. The first method is related to the construction of an artificial neural network in the manual mode. An algorithm is presented that reflects the iterative process of its training. In the second case, the network is built automatically by applying a genetic algorithm. At the beginning of the work, a set of randomly generated initial data arrives at the input of the algorithm. In the course of its work, the genetic algorithm determines the architecture and parameters of the neural network, which ensure the successful solution of the assigned applied problem. Trained networks are used to classify data. Both networks showed acceptable classification accuracy of the results obtained in the course of the students’ knowledge testing.

Purpose: to develop a software technology based on artificial intelligence methods aimed at analyzing large volumes of optical coherent tomography (OCT) data in order to identify early symptoms of age-related macular degeneration (AMD) and the transition from dry AMD to neovascular AMD (nAMD).Material and methods. Patients with dry AMD (1125 eyes), wet nAMD (1200 eyes) and subjects without ophthalmic pathology (1205 eyes) underwent a standard ophthalmological examination and macular OCT (Cirrus HD-OCT 4000, Carl Zeiss Meditec AG, Germany) according to the MacularCube scanning protocol with a standard ETDRS macular map. The thickness of the retina from the inner limiting membrane to the retinal pigment epithelium and the presence and location of fovea configuration changes were determined.Results. Two ultra-precise artificial neural networks (ANN) were created: one intended to identify patients with early signs of dry AMD and the other to identify those with signs of nAMD. By the time when 1000 training images were processed, the image interpretation decision-making accuracy in the first ANN increased to reach 97.6%, in the second ANN, to 96.8%, which shows the high efficiency of this technology. The deep learning processes were run on an Amazon Web Service EC2 GPU. The trained ANN model was also tested on healthy eyes at each estimated probability value. The correspondence between the doctor's diagnosis and the decision taken by the ANN was assessed.Conclusions. A software technology based on artificial intelligence methods has been developed, which answered the need to process a large amount of OCT data. The technology proved effective in identifying early symptoms of dry and wet forms of AMD and in early diagnosis of the transition of the dry form of AMD into neovascular form, the latter requiring immediate treatment.

Purpose of the study. The aim of the study is to create neural network models of modules in an adaptive testing system to design an individual testing trajectory.The research article discusses the implementation of an adaptive testing system in terms of introducing artificial neural network modules into its composition, which should solve the problem of choosing a topic and the complexity of the next question, taking into account previous answers and the complexity of previously asked questions, as well as the connectivity of topics and response time as a factor guessing or searching for an answer, thereby forming an individual testing trajectory.Materials and methods. In the course of the study, the data that affect the quality of the solution of the problem was analyzed, the general modular structure of the system was proposed, and the main data flows entering the input of an artificial neural network (ANN) were described. To solve the problem of choosing the complexity of a question, it is proposed to use a feed-forward network, a comparison of various ANN architectures and training parameters (weight update algorithms, loss functions, number of training epochs, packet sizes) is carried out. As an alternative, the possibility of using a recurrent ANN LSTM (Long-Short Term Memory) network is considered. All results were obtained using the high-level Keras library, which allows you to quickly start at the initial stages of research and get the first results. SGD, Adam, NAdam and RMSprop implemented in Keras were compared as optimizers to achieve faster convergence. Adam showed the best results in terms of accuracy, while the MSE loss function (mean square error) was used together with the optimizer. Traditionally, training was carried out for a large number of epochs; graphs of dependences of accuracy on the number of epochs for a different number of neurons in the hidden layer were experimentally obtained.Results. Based on the study, we can conclude that the obtained accuracy of the direct propagation network of 80-85% is quite sufficient for its use in the adaptive testing system. However, it remains to answer the question of the need to improve the efficiency of an already implemented network, and, therefore, to conduct research on methods to improve the efficiency of networks, including finer tuning of parameters and learning algorithms, as well as architecture.A well-known and obvious drawback of using LSTMs is their exactingness in terms of equipment and resources, both during training (the training process takes a significant amount of time) and during startup, in our case, it is supplemented by increased requirements for the training sample and casts doubt on the advisability of further study of LSTM networks when solving this task.Conclusion. The introduction of the proposed tools will allow implementing an adaptive testing system, with an intelligent selection of questions depending on the demonstrated level of knowledge of the test person to form an individual testing trajectory in order to determine the reliable level of knowledge of the test subject for the optimal number of questions asked.

Recently, there has been an increased interest in the use of artificial neural networks in various branches of the electric power industry including relay protection. The operation of the traditional microprocessor-based relay protection device is based on calculation the RMS values of the monitored current and voltage signals and its comparison with the predetermined thresholds. However, calculated RMS values often do not reflect the real processes occurring in the electrical equipment under protection due to, for example, current transformer saturation. In this case secondary current has a characteristic distorted waveform, which is significantly differs from its ideal (true) waveform. This causes underestimation of the calculated RMS value of the secondary current compared to its true value; also, it causes a trip time delay or even to a relay protection devices operation failure. In this regard, one of the perspective applications of the artificial neural network for the relay protection purposes is the current transformer distorted secondary current waveform restoration due to its saturation. The article describes in detail the stages of the practical implementation of the artificial neural networks in the MATLAB-Simulink environment by the example of its use to reconstruct the distorted secondary current waveform of the saturated current transformer. The functioning of the developed neural networks was verified in the MATLAB-Simulink environment; with the use of the SimPowerSystems component library a model was implemented which allow simulating the current transformer saturation, accompanied by the secondary current waveform distortion, and its further restoration using developed artificial neural networks. The obtained results confirmed the ability of the neural networks that had been developed to almost completely restore the distorted secondary current waveform. Thus, it seems promising to use pre-trained artificial neural networks in real relay protection devices, since such use will ensure the speed of real relay protection devices; their operation reliability will also increase.

The article considers the problem of approximation of the solution of the Cauchy problem for an ordinary differential equation of the second order. The approximation scheme is based on the Taylor expansion of the solution with a remainder in the Lagrange form. The residual term is sought in the form of the output of a neural network with radial basis functions (RBF networks). An algorithm for learning (choosing the optimal parameters) of the RBF network is presented. Using the example of solving the Cauchy problem for a second-order nonlinear differential equation – the Duffing oscillator, the influence of various radial-basis functions on the quality of interpolation and extrapolation of the solution to the Duffing equation is investigated.

The paper is devoted to studying the possibility of using artificial neural networks (ANN) to estimate ground subsidence caused by underground mining. The experiments showed that the most suitable network structure is a network with three layers of perceptrons and four neurons in the hidden layer with the back propagation algorithm (BP) as a training algorithm. The subsidence observation data in the Mong Duong underground coal mine and other parameters, including: (1) the distance from the centre of the stope to the ground monitoring points; (2) the volume of mined-out space; (3) the positions of the ground points in the direction of the main cross-section of the trough; and (4) the time (presented by cycle number), were used as the input data for the ANN. The findings showed that the selected model was suitable for predicting subsidence along the main profile within the subsidence trough. The prediction accuracy depended on the number of cycles used for the network training as well as the time interval between the predicted cycle and the last cycle in the training dataset. When the number of monitoring cycles used for the network training was greater than eight, the largest values of RMS and MAE were less than 10 % compared to the actual maximum subsidence value for each cycle. If the network training was less than eight cycles, the results of prediction did not meet the accuracy requirements.

Relevance of the research topic. An important trend of the development strategy of the agricultural sector`s labour potential is the establishment of the environment for its formation, which fits sustainable development goals. Human factors of production in the context of sustainable development are characterized by the growing role of education and intangible capital. It acts as a determinant of the national economy`s and economic sectors` growth to ensure life conditions of the present and future generations. Rural areas become a kind of framework for residence, demographic trends and proportions, social interaction, gaining psychophysical, motivational and, to some extent, intellectual parameters for harmonious socio-ecological and economic, i.e. sustainable, development. Thus, the issue of levers` assessment of labour potential management in agricultural sector remains relevant at the present stage of achieving sustainable development goals. Problem statement. A sustainable development paradigm is a combination of economic, social and environmental components represented by a significant number of interconnected factors. Their comprehensive impact determines the ways and dynamics of achieving sustainable development goals. Making managerial decisions referring to the impact on each separate determinant and taking into account their interaction is accompanied by the analysis and processing of a significant set of indicators and requires special data processing methods. Solving this problem is possible when applying artificial neural networks to define parameters, which will be studied in the model. Analysis of recent research and publications. Both Ukrainian (M. Artyukh, L. Voynich, O. Lytvyn, I. Lyashenko, I. Tkach, L. Khrushchev, S. Shumska and V. Yankovy) and foreign scholars (N. Kuzina, S. Pshenichnikova and N. Suvorov), and others [1-6] analyzed the possibility of applying economic and mathematical models for economic growth analysis. The results of our own research on the problem are given in [10-16]. Works written by S. Bogachev, O. Galushko, M. Zgurovsky, S. Kvitka, S. Kozlovsky, J. Kologrivov, K. Rulitskaya, O. Sadovnik, L. Fedulova and A. Shpykulyak demonstrate their attempts to apply economic and mathematical methods and models for forecasting the agricultural sector`s sustainable development. Nevertheless, they did not receive further development and practical use for labour potential management in the context of sustainable development. Selection of unexplored parts of the general problem. Despite some progress in problem solving, there is a need for in-depth study of new approaches aimed at forecasting and modeling the agricultural sector`s targets based on the threefold concept of sustainable development using modern management methods, taking into account environment dynamism and the complexity of positive and negative externalities. Problem statement, research goals. The above-mentioned circumstances determine the rationale to assess the levers of influence over the target sustainable development indicators in the agricultural sector. Method and methodology of research. When doing the research general scientific (analysis and synthesis, abstract-logical, generalization and system analysis) and special (generalized linear regression, artificial neural networks) research methods have been used. Presentation of the main material (results of work). The neural network modelling allowed building a multifactor impact model on the resulting indicator of sustainable development in the agricultural sector, namely labour productivity. The next factors have been identified in the model: average monthly nominal wage; energy security; power-weight ratio; number of trucks (per 100 hectares of sown area and per 1,000 employees); amount of mineral fertilizers applied; amount of applied organic fertilizers for crops; emissions into the atmosphere from stationary and mobile sources of pollution; total waste accumulated at landfills; average index of regional human development; capital investment per capita. The following parameters like average monthly nominal wage, application of mineral fertilizers, number of trucks, and capital investments significantly affect the modelling results. The proposed model allows modelling and forecasting, based not only on previously obtained indicators and their dynamics, but also, which is important in the context of sustainable development, to set targets which facilitates managerial influence not only on the end result but on the process if its achievement as well, including the optimization impact. In addition, the modelling allows to adjust the impact factors, if they are either insignificant, as it has been found out when modelling, or lose value due to technological changes (for example, energy security and power-weight ratio). Conclusions. Based on the proposed multifactor adaptive model of sustainable development in the national agricultural sector, the strategic trends for its labour potential management have been developed. The neural network modelling methodology applied when building a model provided an opportunity to take into account the tasks for long-term sector`s sustainable development while determining and dynamically adjusting priorities, parameters and levers of influence on current industry`s efficiency. The proposed model allows not only modelling and forecasting based on previously obtained indicators and the dynamics of their adjustment, but also to set targets to obtain a number of possible scenarios for system development. It depends on forecasting conditions and parameters. It does not only rise the validity of managerial decisions, but also ensures the timeliness of management object`s adaptation to the ever-changing environmental conditions. Moreover, it provides managerial influence not only on the result, but also on the process of its achievement, including optimization of sustainable development levers.

The relevance of the study is caused by the significantly increased requirements for control systems for surface facilities of oil and gas fields. The control systems should ensure the safe conduct of processes while maintaining the specified quality of commercial oil in the presence of an intense background of disturbances, the ability of long-term autonomous functioning with minimal impact from operating personnel, minimization of the risks of negative impact of production factors on the environment, etc. A prerequisite for meeting the above requirements and a way to increase the level of intellectualization of control systems for oil and gas treatment facilities is the development of improved control systems using methods of mathematical modeling, automatic control theory and information technologies. The goal of the research is to develop an integrated mathematical model for building an intelligent control system for processes occurring in one of the main devices of an oil treatment plant – a three-phase separator. The object of the research is a three-phase separator, which includes sections for degassing, dehydration and unloading of oil emulsion. Methods: Katz's method for calculating the component composition of the gas and liquid phases of a separating oil emulsion, material and molar balances, methods for the numerical solution of nonlinear equation systems, configuration and training of artificial neural networks MATLAB (NFTool, NNTool), principles of cell modeling, block modeling using MATLAB Simulink software. Results. The author has developed a complex mathematical control model that describes the statics and dynamics of processes occurring in each section of a three-phase separator. To identify the real process of emulsion dehydration, the intensity function – the analogue of the laboratory sedimentation curve, was proposed. The structure of the separator cell model is parametrically related to the intensity function. To solve a general problem of dynamics in a single cycle, which includes several systems of nonlinear equations, local iterative procedures are excluded from the cycle. Blocks of artificial neural networks trained for iterative solution of local nonlinear systems are added. The model was configured using the MATLAB Simulink software. A solution is obtained, illustrated by graphs of static dependencies and transient characteristics of a three-phase separator in terms of its main input parameters. Conclusion. The obtained results of mathematical modeling of a three-phase separator, as well as the results of modeling an oil emulsion heater previously published by the author, allow us to consider the developed models as a basis for building an improved control system for oil treatment facilities in an intelligent oil and gas field.

Recently, there has been an increased interest in the use of artificial neural networks in various branches of the electric power industry including relay protection. Аrtificial neural networks are one of the fastest growing areas in artificial intelligence technology. Recently, there has been an increased interest in the use of аrtificial neural networks in the electric power engineering, including relay protection. Existing microprocessor-based relay protection devices use a traditional digital signal processing of the monitored signals which is reduced to a multiplying the values of successive samples of the monitored current and voltage signals by predetermined coefficients in order to calculate their RMS values. In this case, the calculated RMS values often do not reflect the real processes occurring in the protected electrical equipment due to, for example, current transformer saturation because of the DC component presence in the fault current. When the current transformer is saturated, its secondary current waveform has a characteristic non-periodic distorted form, which is significantly differs from its primary (true) waveform, which causes underestimation of the calculated RMS value of the secondary current compared to its true value. In its turn, this causes to a trip time delay or even to a relay protection devices operation failure. The use of аrtificial neural networks in conjunction with a traditional digital signal processing provides a different approach to the functioning of both the measuring and logical parts of the microprocessor-based relay protection devices, which significantly increases the speed and reliability of such relay protection devices in comparison with their traditional implementation. A possible application of the аrtificial neural networks for the relay protection purposes is the fault occurrence detection and its type identification, current transformer secondary current waveform distortion restoration due to its saturation up to its true value, detection the distorted and undistorted sections of the current transformer secondary current waveform during its saturation, primary power equipment abnormal operating modes detection, for example, power transformer magnetizing current inrush. The article describes in detail the stages of the practical implementation of the аrtificial neural networks in the MATLAB-Simulink environment by the example of its use to restore the distorted current transformer secondary current waveform due to saturation.

Introduction. Artificial neural networks are a popular tool of contemporary research and technology, including food science, where they can be used to model various technological processes. The present research objective was to develop an artificial neural network capable of predicting the content of isogumulone in a hop extract at given technological parameters of the rotary pulse generator. Study objects and methods. The mathematical modeling was based on experimental data. The isogumulone content in the hop extract I (mg/dm3) served as an output parameter. The input variables included: processing temperature t (°C), rotor speed n (rpm), processing time  (min), and the gap between the rotor teeth and stator s (mm). Results and discussion. The resulting model had the following parameters: two hidden layers, 30 neurons each; neuron activation function – GELU; loss function – MSELoss; learning step – 0.001; optimizer – Adam; L2 regularization at 0.00001; training set of four batches, 16 records each; 9,801 epochs. The accuracy of the artificial neural network (1.67%) was defined as the mean relative error. The error of the regression model was also low (2.85%). The neural network proved to be more accurate than the regression model and had a better ability to predict the value of the output variable. The accuracy of the artificial neural network was higher because it used test data not included in the training. The regression model when tested on test data showed much worse results. Conclusion. Artificial neural networks proved extremely useful as a means of technological modeling and require further research and application.

Management of solid organ recipients requires a significant amount of research and observation throughout the recipient’s life. This is associated with accumulation of large amounts of information that requires structuring and subsequent analysis. Information technologies such as machine learning, neural networks and other artificial intelligence tools make it possible to analyze the so-called ‘big data’. Machine learning technologies are based on the concept of a machine that mimics human intelligence and and makes it possible to identify patterns that are inaccessible to traditional methods. There are still few examples of the use of artificial intelligence programs in transplantology. However, their number has increased markedly in recent years. A review of modern literature on the use of artificial intelligence systems in transplantology is presented.

Breast cancer is a leading cause of death in women due to cancer. According to WHO in 2018, it is estimated that 627.000 women died from breast cancer, that is approximately 15 % of all cancer deaths among women [3]. Early detection is a very important factor to reduce mortality by 25-30 %. Mammography is the most commonly used technique in detecting breast cancer using a low-dose X-ray system in the examination of breast tissue that can reduce false positives. A Computer-Aided Detection (CAD) system has been developed to effectively assist radiologists in detecting masses on mammograms that indicate the presence of breast tumors. The type of abnormality in mammogram images can be seen from the presence of microcalcifications and the presence of mass lesions. In this research, a new approach was developed to improve the performance of CAD System for classifying benign and malignant tumors. Areas suspected of being masses (RoI) in mammogram images were detected using an adaptive thresholding method and mathematical morphological operations. Wavelet decomposition is performed on the Region of Interest (RoI) and the feature extraction process is performed using a GLCM method with 4 statistical features, namely, contrast, correlation, entropy, and homogeneity. Classification of benign and malignant tumors using the MIAS database provided an accuracy of 95.83 % with a sensitivity of 95.23 % and a specificity of 96.49 %. A comparison with other methods illustrates that the proposed method provides better performance.

The paper considers the development of a method for positioning a mobile device using a sensor network of BLE-beacons, the approximation of RSSI values and artificial neural networks. The aim of the work is to develop a method for positioning small-scale industrial mechanization equipment for building unmanned systems for product movement tracking. The work is divided into four main parts: data synthesis, signal filtering, selection of BLE beacons, translation of the RSSI values into a distance, and multilateration. A simplified Kalman filter is proposed for filtering the input signal to suppress Gaussian noise. A description of two approaches to translating the RSSI value into a distance is given: an exponential approximation function with a coefficient of determination of 0.6994 and an artificial feedforward neural network. A comparison of the results of these approaches is carried out on several test samples: a training one, a test sample at a known distance (0–50 meters) and a test sample at an unknown distance (60–100 meters). The artificial neural network is shown to perform better in all experiments, except for the test sample at a known distance (0–50 meters), for which the r.m.s. error is higher by 0.02 m 2 than that for the approximation function, which can be neglected. An algorithm for positioning a mobile device based on the multilateration method is proposed. Experimental studies of the developed method have shown that the positioning error does not exceed 0.9 meters in a 5×5.5 m room under monitoring. The positioning accuracy of a mobile device using the proposed method in the experiment is 40.9 % higher. Experimental studies are also conducted in a 58.4×4.5 m room, showing more accurate results compared to similar studies.

In this paper we solve the problem of synthesizing a power regulator in a traction electric drive system using artificial neural networks. To control the vehicle and obtain the desired quality of transients, neural network observers have been developed that allow the measurement of indirect parameters to determine the immutable coordinates of the system. For this purpose, this paper uses dynamic neural networks. When developing the neural network observer, experimental data obtained by the authors on an operating vehicle in real operating conditions are used. To test the effectiveness of using the created artificial neural network, an object is simulated with a random nature of the supply voltage change. A comparative analysis of transients in a system with a power neuroregulator and classical regulators in a subordinate control system shows a fairly high convergence of the results.

The relevance. These are the first studies in the Kivi region. Due to the presence of titanium and zinc in the area, these studies are necessary. Artificial Neural Network and K-means methods for element behavior measurement are new methods in mineral exploration. The main aim of the research is to identify Ti and Zn geochemical behavior for prediction Ti by ANN and K-means methods. Object: Kivi 1:100000 geochemical map in Ardabil province, Iran. Methods. The samples taken from bottom sediments of the Kiwi region, which were analyzed by the ICP-MS method, served as the initial data. Then, the behavior of these elements in relation to each other and their geographical coordinates was analyzed by the K-means clustering method. The amount of titanium was also predicted with the artificial neural network (ANN- GRNN). Results. The Ti and Zn elements relationship was determined using this K-means method taking into account the latitude and longitude of the samples to estimate the grade and more accurate estimation of the appearance and extent of the geochemical halos in the studied area. According to the results obtained during processing of these elements, a regression equation was drawn up to estimate the titanium content based on three parameters: Zn content, the length and width of the sampling points, the correlation coefficient. According to the K-means cluster centers and artificial neural network, the Ti element grade was predicted and the correlation coefficient was reported 0,51. Both methods produce the desired results, but the artificial neural network method has more accurate data. Schematic maps of the initial and predicted Ti content were constructed. The results of the study can be used in the course of geological exploration to forecast and identify new promising areas.

The article is devoted to the problem of improving the accuracy of short-term load forecasting of electrical engineering complex of regional electric grid with the use deep machine learning tools. The effectiveness of the application of the adaptive learning algorithm for deep neural networks for short-term load forecasting of this electrical complex has been investigated. The issues of application of convolutional and recurrent neural networks for short-term load forecasting are considered. A comparative analysis of the accuracy of the short-term load forecasting of electrical engineering complex of regional electric grid obtained using the ensemble neural network method and single neural networks are produced.

Background. Scientific article reveals the problem of analyzing, recognizing and managing highdefinition images with a minimum error due to the previous frame-by-frame recognition of a complex of lowdefinition images. The fundamental problem is the appearance and impact of gradient noise in the form of disaggregated pixel segments, which significantly reduce the resolution of the area under consideration. Materials and methods. Until now, this area of research on artificial neural networks has not been sufficiently studied due to low consumer demand for the technology and slow development from enthusiasts. Despite the fact that image recognition was not a promising direction before, at the moment it holds potential in the field of application of artificial neural networks and gradient noise leveling with deep learning based on them. Results and conclusions. The article considers both the possibility of adapting old existing approaches to solving the problem of pattern analysis and recognition, and a new control method based on a complex of storyboarding artificial neural networks with further integration for deep learning and solving problems.

THE PURPOSE. Improving the accuracy of forecast calculations of demand for energy resources is an urgent task, especially in the light of the Digital Energy of the Russian Federation program. Prediction is also required for he at supply systems. The complexity of the analysis is the lack of confirmation of the similarity properties of energy systems and complexes for buildings with similar functionality. On the example of buildings of secondary educational institutions located i n the territory of Moscow, the assumption of heteromorphism of thermal systems is proved. METHODS. In the work, an assumption was made that there were no significant changes in the data on the heat consumption of the energy facilities of schools, which was confirmed by the absence of changes in the average annual heat consumption and jumps in the monthly heat consumption diagrams. The amount of heat energy consumption measured and transferred to the IS is influenced by a number of additional factors: accura cy drift of heat energy metering devices; aging and overgrowing of the internal surfaces of the building's heating network equipment; physical aging and deterioration of the building envelope and deterioration of their thermal insulation performance. When compiling predicted energy consumption, this means that it is permissible to use not only statistical data about the analyzed object itself, but also about a variety of objects similar to those analyzed in structure and functionality. RESULTS. A set of input factors is proposed that makes it possible to accurately determine the predicted demand for thermal energy for buildings of secondary educational institutions. The possibility and similar accuracy of the results of forecasting the demand for thermal ene rgy is shown both through the use of multivariate regression analysis and artificial neural networks. CONCLUSION. ЭBased on the combined use of various mathematical approaches, it is proposed to use the methodology for forecasting energy demand by energy complexes and systems as a mechanism for determining the correctness of the transmitted meter readings.

This article presents the development and study of a model for formalizing the diagnosis process using artificial intelligence methods. At present, various artificial neural networks and expert systems have been created and are used in the diagnosis. The analysis of these works showed that these methods show good results, but they have a number of disadvantages, the most significant of which is the complexity of the organization and the large time spent on training the neural network. Thus, the problem is posed of developing new algorithms that have the probability of making an accurate diagnosis, comparable to artificial neural networks and expert systems, and at the same time having less training time. One of the ways to solve this problem is to develop a model for diagnosing diabetes mellitus based on an artificial immune system. The aim of the work is to develop and study a model for formalizing the diagnosis process using artificial intelligence methods. A model of the diagnosis process is considered: pre-diabetes state (impaired glucose tolerance, impaired fasting glucose), type I diabetes, type II diabetes. The problem of diagnosing a disease can be viewed as a classification problem. In this work, the process of making a diagnosis was considered as dividing these analyzes and anamnesis of patients into four classes corresponding to one of the diagnoses: pre-diabetes state (impaired glucose tolerance, impaired fasting glycemia), type I diabetes, and type II diabetes. To solve this problem, an artificial immune system and an artificial neural network of Kohonen were used. An artificial immune system represents an idealized version of a natural analogue and reproduces the key components of a natural process: selection of the best antibodies in a population depending on the degree of their affinity (proximity) to an antigen, cloning of antibodies, and mutation of antibodies.