Software & Systems

Articles of journal № 2 at 2025 year.

Order result by: Public date | Title | Authors

1. Anytime algorithms for intelligent real-time decision support systems (in terms of a pattern classification task) [№2 за 2025 год]
Authors: Eremeev, A.P. , Bashkova, S.M.
Visitors: 2744
A relevant task for developing methods and software tools for designing modern and prospective intellectual systems of real-time decision support operating under fairly strict time constraints is the development of flexible (anytime) algorithms for decision search. Such algorithms are capable of producing an acceptable solution from a certain point in time, gradually improve it until the optimal solution is obtained with a corresponding increase in computer resources (usually time). The aim of this work is to study and develop flexible decision search algorithms to apply them in real-time intelligent decision support systems under strict restrictions on the response time when problematic (emergency, abnormal, etc.) situations arise. The authors consider algorithms based on the neural network approach on the example of solving the problem of pattern (image) classification. They give a comparative analysis of neural networks using the early output method to solve such problem. They propose an original anytime modification of the neural network that allows obtaining an earlier solution for pattern (image) classification compared to the classical approach. This is relevant for real-time systems. The authors describe the software implementation based on the Tensorflow and Keras frameworks of the Python programming language, as well as the results of computer modeling, which confirm that the proposed approach is promising. The authors carry out research and development in terms of creating basic tools for building intelligent real-time decision support systems to help operational and dispatch personnel (decision makers) in controlling, monitoring and diagnosing complex technical and organizational systems, as well as in recognizing and classifying problem situations.

2. Applying a genetic algorithm to determine the set of enemy airborne weapon [№2 за 2025 год]
Authors: Akodit, E.V., Rybalchenko, P.V.
Visitors: 2352
Combat operation modeling requires substantial time expenditures. Moreover, modeling should be multivariate to ensure the adequacy and reliability of the calculations performed. The analysis of computational and analytical activities of command and control bodies showed a significant increase in the labor input of the military simulation complex operator. The authors found that most of the time allocated to simulation modeling is spent on inputting the initial data that define air enemy action scenarios. They proposed an approach to solve this problem by building an intelligent model that provides automated input of input data for an enemy. Thus, the entered data should meet the requirements of adequacy, and the intellectual model itself should meet the requirements of reliability. For this purpose, it is necessary to determine the rational nature of action from the enemy's point of view. The paper considers one of the most important tasks – determining the nomenclature and number of air assault vehicles participating in a strike. The authors propose a method of automated determination of air assault vehi-cles in a strike based on a genetic algorithm apparatus. The developed fitness function for the genetic algorithm considers the damage to a given set of defense objects in a strike, as well as strike cost. The authors define the rules for initialization and crossing of individuals (set of actions of air assault vehicles), rules for their selection to form the next generation population. Using a certain example, they consider the conditions for solving the problem and determine the computational complexity of determining air assault vehicles. The authors obtained and described the application results of the proposed method in simulation modeling.

3. Ontology synthesis for decision support systems based on large language models [№2 за 2025 год]
Authors: Borisov V.V., Misnik, A.E., Sheroburko, E.N., Khabarov A.R.
Visitors: 2196
The paper focuses on automated forming of ontologies for management and decision support systems using large language model technologies. With exponential growth of data volumes and increasing complexity of information systems, traditional ontology development methods are becoming less effective due to their labor-intensive and time-consuming nature. The au-thors propose a methodology that allows automating the processes of extracting, analyzing and structuring knowledge about the subject area. This is especially relevant for tasks with a large amount of multistructured information. The authors pay special attention to the integration of language models with ontological structures to form concepts, their attributes and relationships based on textual data. The methodological basis of the research includes natural language processing algorithms, transformer architectures and meta-associative graphs. This allows not only extracting knowledge from texts, but also formalizing it in a form convenient for further processing. Transformers ensure high accuracy in analyzing texts and identifying key concepts. Meta-associative graphs allow visualizing and efficiently integrating heterogeneous data. An important feature of the proposed approach is the ability to update ontologies dynamically as new data becomes available. This increases the relevance and accuracy of the generated models. The main research results include a developed methodology for automated ontology generation and it is testing on the example of public procurement planning tasks. The authors prove the effectiveness of using language models for open-source analysis and automated ontology formation that allow optimizing procurement procedures and increasing the efficiency of resource allocation. The practical significance of the research is creating a tool capable to support management decision making, automating data structuring processes and minimizing a human factor effect. Using the methodology contributes to reducing time costs, improving the analysis accuracy and forming more flexible and adaptive knowledge management systems.

4. Modeling and visual analysis of vortex flows in computational fluid dynamics [№2 за 2025 год]
Author: Goryachev, V.D.
Visitors: 2866
Visual analysis and interpretation of gas-dynamic structures are necessary components of numerical modelling of technological processes and natural phenomena. The paper discusses methods and work features in the system of visualization and interpretation of numerical simulation results of graphical post-processing of calculations on high-performance computers. The system environment enables to carry out detailed visual analysis of scalar, vector and tensor fields of gas and liquid flows obtained during modeling of power engineering devices and analyzed during digital description of natural processes. A particular feature of work in the system is variable processing of modeling results with combined graphical representation of primary fields (velocity, pressure, temperature, concentration). Additional generation of secondary, derived quantities, tensor invariants of velocity gradients and other field characteristics is also performed. The in-depth analysis is aimed at visualizing the coherent flow structures in a more visual and representative form. The authors use a combined approach to highlighting the hidden relationships of flow characteristics through the reflection of symbolic derived fields defined in the analytical dependence editor in video scenes to show flow features. The functional relationship between the fields is defined by activating different methods in the basic graphical pipeline of a visualization scene. The authors discuss the methods of expressive visualization of coherent flow structures on the example of two numerical modeling problems of different scale flows. The first problem concerns determining secondary vortex currents formed during the gas turbine blade flow, and the second one concerns modeling the temporal evolution of gas nebula structures. The computational astrophysics problem models the vortex transformation of a system of molecular clouds after their collision with a powerful shock wave from a supernova explosion. The authors compare the methods of visual representation of coherent structures and multiscale turbulence in the analyzed flows with known techniques of postprocessing in computational fluid dynamics.

5. Quantum soft computing technology in the software-algorithmic platform of robust intelligent control system for a robotic manipulator [№2 за 2025 год]
Authors: Borovinsky, V.V., Nikolaeva A.V., A.G. Reshetnikov, Ulyanov, S.V.
Visitors: 2590
The paper considers a method of coordination management for partitioned knowledge bases using quantum soft computing technologies. The method is implemented using soft computing-based knowledge base optimizer SCOptKBTM. The authors apply the control decomposition method, which assumes that each fuzzy controller with embedded knowledge base controls one link of the control object. Coordination control in an Intelligent Control System is enabled through extracting of quantum information about the interrelationships of existing fuzzy controllers for three links of the manipulator with knowledge bases obtained for regular control situations. This system is based on a soft computing technology with partitioned control. For this purpose, a generalizing link that is a quantum fuzzy inference model becomes a part of the intelligent superstructure block. The authors conducted numerical and physical experiments to compare the performance of intelligent control system on quantum computing knowledge base optimizer (QCOptKBTM) with an intelligent control system on soft computing knowledge base optimizer with partitioned control. They proved that the overall control quality score is higher in terms of an intelligent control system on a knowledge base optimizer on quantum computing (for spatial, spatiotemporal and temporal correlations). This is a consequence of introducing into the intelligent control system structure an additional quantum fuzzy inference link that organizes coordination control. The authors demonstrate the robustness of intelligent fuzzy controllers on the example of a self-organizing intelligent control system of a locally unstable and significantly nonlinear controlled object. They show that intelligent control allows guaranteed achieving control objectives in unpredictable management situations and with minimal resource consumption.

6. Constructing a subject OWL ontology: Comparing the effectiveness of different text data mining algorithms [№2 за 2025 год]
Authors: Dyrnochkin, A.A., Moshkin, V.S., Yarushkina, N.G.
Visitors: 3236
The paper describes a software service for analyzing textual information in order to form OWL-ontology when analyzing the state of complex technical systems, for example, the subject domain of oil production systems that output the real-time data. The main direction of the service is processing textual information to form OWL-ontology. OWL-ontology is a formal descriptive model that allows semantical structuring and formalizing information. The paper presents the basic principles of this service functioning. The software service performs preprocessing of textual data, including tokenization, stop word removal, lemmatization and key term extraction. These operations are necessary to improve the quality and integrity of data before further analysis. The main stage of the service is to extract keywords and terms from texts using machine learning algorithms and statistical methods. The service groups texts by similarity and forms clusters. This stage allows revealing a hid-den structure in textual data and identifying common themes or concepts. The described experimental results confirm that the software service is effective for forming ontological structures. They demonstrate the ability of the service to classify and group text data, which is an important step in the process of creating semantic models of complex technical systems. The paper also provides a comparative study of three different concepts of keyword and term extraction: statistical keyword extraction methods, clustering and topic modeling. The authors evaluate the performance of each method based on keyword extraction accuracy, relatedness and understandability of key terms. The study results conclude the advantages and disad-vantages of each approach. In addition, they allow determining the most productive method for specific tasks of ontology building in information systems.

7. Artificial neural network technologies for building personalized mobile application interfaces [№2 за 2025 год]
Author: Tagirova, L.F.
Visitors: 1931
The paper focuses on automating information processes of forming personalized mobile application interfaces on the example of training systems. The authors propose an algorithm for adapting the values of interface components based on artificial neural network technologies. The research subject is mathematical methods, algorithms, software for personalizing interfaces of mobile applications of training systems. The authors use an artificial neural network as a tool for selecting values of interface components. Input data in the network operation are distinctive features of users and technical characteristics of a mobile device. The output data are the values of interface components. The main result of the research is a two-stage algorithm for mobile application interface adaptation. The algorithm is implemented in the developed mobile application of a training system. The conducted research has practical significance since the implementation of the developed mobile application can provide a comfortable interaction between a student and the application. This will contribute to improving the effectiveness of student training.

8. Automated drone image making using pre-trained artificial intelligence models [№2 за 2025 год]
Authors: Bessarabov N.A., Sapozhnikov, A.A., Tatarnikov, D.V., Tyugunov, R.R., Tsyganov, A.M.
Visitors: 2082
The paper discusses an image-mapping method for drone images using pre-trained artificial intelligence models for subsequent training of different class object detectors. The approach consists of three main steps: image segmentation, attention zone extraction, and verifying the segments for the target class using a majority committee of multimodal models. The authors use the Segment Anything Model for object segmentation. Since there can be a significant number of segments, the authors propose an algorithm for building a hierarchical structure over a set of segments. It means that if a segment is in another segment, then this segment will be nested in the segment containing it. Further processing has the following logic for significant acceleration: if the target class is not found in the parent segment, it will not be found in the nested segments. The authors propose an algorithm for finding attention zones as the most probable areas for finding target class objects in the second stage. For this purpose, they use the pretrained SigLIP model. All segments from the processed image set are ranked by decreasing probability of finding the target class in them using attention zones. The authors apply the majority committee of three multimodal chatbots (LLaVA, CogVLM, and Mini-Gemini) in the third stage. There is a stopping rule for the chatbots to check, to avoid enumerating all segments and thus speed up the image partitioning process. The authors conducted a computational experiment to demonstrate the proposed approach metrics and speed.

9. QAMODEL software package: Computer simulation of high-frequency geoacoustic emission [№2 за 2025 год]
Authors: Sergienko, D.F., Parovik, R.I.
Visitors: 1707
The authors propose the QAMODEL software package for quantitative and qualitative analysis of the mathematical model of high-frequency geoacoustic emission. The mathematical model is a coupled system. It consists of two linear ordinary differential equations of the second order with non-constant coefficients and corresponding initial conditions (Cauchy problem). This model describes the interaction of two linear oscillators using linear coupling between radiation sources of high-frequency geoacoustic emission. The authors performed quantitative analysis of the emission mathematical model using the numerical Rosenbrock method of the fourth accuracy order. It is sufficiently robust to the stiffness of the system under consideration. The numerical method is implemented in the QAMODEL software package, which also enables visualization of the simulation results. In particular, the authors used the possibility of plotting oscillograms and phase trajectories at differ-ent values of model parameters entered by the user through the software package interface. It is possible to save the plots with png extension for further study. The qualitative analysis of the mathematical model of high-frequency geoacoustic emission involved studying the stiffness property. The QAMODEL software package is able to construct a stiffness function over time at different values of mathematical model parameters of high-frequency geoacoustic emission. The QAMODEL software package is written in C++ using Qt cross-platform software development framework. The software package was registered in Rospatent.

10. Increasing performance of the radiation transport simulation: TPT3 parallel program [№2 за 2025 год]
Authors: Galyuzov, A.A., Kosov, M.V.
Visitors: 1837
The problem of increasing performance of radiation transport modeling programs is relevant for detailed calculations of radiation effect on nuclear facility structural elements. One of the ways to increase computational performance is to use multicore CPUs with vector instructions. Programs with sequential code, such as MCNP and Geant4, are able to parallelize only events initiated by independent particles, their transport simulation is performed sequentially, which does not allow full using of multicore CPU capabilities. The high-performance program TPT3 for parallel simulation of a radiation transport has been developed in the Dukhov’s Institute VNIIA. It utilizes parallel computations and vectorization capabilities as efficient as possible and running on both central processors and graphics accelerators. This required simplifying the description of the installation geometry to the voxel level and restricting the interaction algorithms to one-in-two reactions, where only two particles are born when a particle interacts with matter. On the other hand, particle transport is not performed independently in TPT3, but as a multi-million weight particle ensemble, which allows combining TPT3 calculations with hydrodynamic algorithms. The paper demonstrates that using TPT3 program in the simulation tasks of an ion cascade and neutron yield in a neutron generator on multicore central processors provides a 1.5-fold performance increase compared to similar pseudo-parallel simulation of the Geant4 program. Graphics accelerators allow increasing the performance by another order of magnitude. In addition, a weighted neutron transport modeling algorithm has been developed in the TPT3 program, which makes it possible to simulate the nuclear fission chain with a characteristic exponential increase in the number of real neutrons while maintaining a limited number of simulated weighted neutrons.

| 1 | 2 | 3 | Next → ►