Software & Systems

1. Сonstructing local fuzzy models for complex object situational control based on precedents

[№3 за 2025 год]
Authors: Dli M.I. (midli@mail.ru) - (Smolensk Branch of the Moscow Power Engineering Institute, Ph.D; Sokolov, A.M. (ansokol98@mail.ru) - NRU “MPEI” (Postgraduate Student), ; Vorotilova, M.Yu. (rita.vorotilova@mail.ru) - Smolensk Branch of the Moscow Power Engineering Institute (Junior Researcher);
Abstract: The paper discusses the issues of optimal control of various nature complex objects in dynamically changing environmental uncertainty factors. The authors analyzed existing control approaches for this category of objects, taking into account specific requirements for developing modern intelligent decision support systems. The paper proved the effectiveness of fuzzy situational-precedent modeling, where past cases form the knowledge base and enable clear visualization of optimal management decisions. The results indicate that model implementation faces adaptation challenges when management conditions evolve, primarily due to the extensive set of incorporated situations and their transition dynamics. Authors present innovative fuzzy situational-precedent models confined to specific feature space domains, designed to represent situational control parameters. The main feature of the proposed models is a limited area of their construction to a certain space around the current and target situations, which greatly simplifies the process of adapting the network structure when internal and external factors change. The authors proposed three alternative approaches for defining the feature space domain and, consequently, determining the network structure complexity of the model based on specifying a localization coefficient. The paper describes a method for constructing local fuzzy models that involves merging similar fuzzy precedents upon reaching target situations, thereby simplifying model structure and enhancing situational control responsiveness for complex objects. The software implementation of the proposed method is made in Python 3.12. using the Numpy library and the NetworkX package for network visualization for fuzzy calculations. The authors conducted a computational experiment demonstrating the effectiveness of local fuzzy situational-precedent models for managing complex objects under dynamic uncertainty compared to fixed-structure situational models.
Keywords: local fuzzy model, situational control, case-based approach, complex object
Visitors: 1307

2. Functional modeling of obstacles for the problem of determining the flight route of an unmanned aerial vehicle

[№3 за 2025 год]
Authors: Chentsov, A.E. (aleks2058@mail.ru) - Zhukov Air and Space Defence Academy (Dr. Sci. (Military Sciences), Senior Researcher); Ishchenko, M.A. (mihamuz@rambler.ru) - Zhukov Air and Space Defence Academy (Cand. of Sci. (Engineering) Lecturer); Pankova, A.E. (Nastya-Pankova1998.1998@mail.ru) - Zhukov Air and Space Defence Academy (Adjunct);
Abstract: The article presents an analysis of methods for determining the flight route of an unmanned aerial vehicle. Four main groups of methods are distinguished: spatial decomposition methods, consisting in dividing the space into many areas; methods based on route networks, based on the use of visibility graphs and on the use of generalized Voronoi diagrams; methods based on potential fields, based on the physical analogy with the movement of a charged particle in an electrostatic field; methods based on artificial intelligence technologies. The advantages and disadvantages of these methods are given in the table. Particular attention is paid to the navigation graph method, which boils down to finding the closest vertices to the start and end points, since this method is the most suitable for the problem of determining the flight route of a UAV. The process of constructing a navigation graph on a plane is presented and the need for its formation in three-dimensional space is substantiated. The functional-voxel method is considered, which is based on the principle of linear approximation of the studied function space for calculating its local geometric characteristics that allow describing a geometric object, and its advantages and practical application are considered. A mathematical apparatus is presented for forming three-dimensional functional models of geometric description of obstacles of artificial (detection, destruction and suppression zones) and natural (terrain relief) origin, which will allow processing the entire three-dimensional space of the studied environment. Based on these functional models, possible areas of flight of an unmanned aerial vehicle are identified and a navigation graph of possible flight routes in these areas is formed.
Keywords: functional model, routing task, unmanned aerial vehicle, military simulation complex, geoinformation computer modeling, navigation graph
Visitors: 1247

3. Multilevel accuracy assessment of prediction tools for a genetic sequence functional structure

[№3 за 2025 год]
Authors: Arzhaev V.I. (arzhaeVI@cps.tver.ru) - R&D Institute Centerprogramsystem (Branch Manager), Ph.D; Skvortsov А.V. (skvortsovAV@cps.tver.ru) - R&D Institute Centerprogramsystem (Head of Department), Ph.D;
Abstract: The paper examines several methodologies for quantitative assessment of functional structure prediction quality in genetic sequences using available gene prediction tools. It focuses on algorithms and software tools for predicting functional structure in genetic sequences. The paper analyzes quantitative prediction accuracy parameters for functional structure in genetic sequences and their algorithmic implementations. The authors explored current development of methods for comparing functional annotation of genetic sequences, as well as methods for predicting a genome functional structure. As a result, they selected quantitative similarity metrics for functional annotation elements in nucleotide sequences. These metrics involve both nucleotide-level resolution and gene exon-intron structure. The calculation methods were adapted to assess the reliability of existing gene prediction software outputs. The methodology is applicable at both nucleotide and exon levels. Using the selected and refined methods for comparing reference and predicted functional sequence structures, the authors developed a software tool for assessing protein-coding gene prediction quality. The paper describes the static architecture of the developed program and a generalized algorithm for generating statistical quality metrics comparing gene predictions against reference annotations. Unlike existing open-source tools, the proposed solution calculates more informative gene prediction accuracy metrics that surpass basic false-positive and false-negative measures.
Keywords: prediction tools, functional structure, genetic sequence, multilevel accuracy, software
Visitors: 1386

4. Synthesis of adaptive control in multichannel navigation receivers based on intelligent pre-correlation signal processing

[№3 за 2025 год]
Authors: M.A. Likhachev (likhachevma@cps.tver.ru) - R&D Institute Centerprogramsystem (Chief of Department); Smirnov, V.D. (smirnovvladimir057@yandex.ru) - R&D Institute Centerprogramsystem (Engineer-Programmer);
Abstract: The paper focuses on a two-stage synthesis of adaptive control solutions for digital antenna array weight coefficients in navigation receivers under intentional noise and imitation interference. The solution employs intelligent spatiotemporal analysis of received pre-correlation satellite navigation signals. Complex electromagnetic environment, when the desired signal level is significantly below the noise and interference floor, makes conventional spatial filtering methods insufficiently effective. The paper proposes a modified two-stage processing algorithm for navigation signals. The first stage forms weights that provide deep dips in the radiation pattern in the directions of active interference sources whose power exceeds the noise level. The second stage generates new weighting coefficients based on detecting spoofing signals and estimating their arrival angles. Spoofing signal source coordinates (azimuth and elevation) are predicted using an LSTM neural network. Training data comprised cyclic autocorrelation matrices encoding the cyclostationary signatures of genuine satellite signals. This is particularly critical in spoofing scenarios where interference mimics authentic navigation signal parameters and cannot be effectively mitigated by conventional countermeasures. The proposed approach to synthesizing intelligent solutions for managing the set of adaptive weighting coefficients of receiving channels minimizes the impact of combined interference. This approach maintains the integrity and authenticity of received navigation signals for subsequent correlation processing and tracking operations. The key research results include the development and simulation of a two-stage adaptive signal processing algorithm, along with comparative performance analysis against conventional methods (MUSIC and DNN). The results demonstrate that the LSTM neural network provides superior accuracy and stability in signal direction-of-arrival estimation under low signal-to-noise conditions, including for angular coordinates outside the training dataset. The practical significance of this paper is in its potential implementation within modern navigation systems to enhance their resilience against intentional interference and ensure reliable operation in challenging electromagnetic environment.
Keywords: adaptive management, intelligent signal processing, LSTM neural network, global navigation satellite systems (GNSS), spoofing, navigation receiver, combined jamming and spoofing interference
Visitors: 1067

5. Anomaly detection in container systems: Frequency analysis and hybrid neural network

[№3 за 2025 год]
Authors: Kotenko, I.V. (ivkote@comsec.spb.ru) - St. Petersburg Federal Research Center of the Russian Academy of Sciences (Professor, Chief Researcher, Head of Laboratory), Ph.D; Melnik, M.V. (mkmxvh@gmail.com) - St. Petersburg Federal Research Center of the Russian Academy of Sciences (Postgraduate Student);
Abstract: Continuously improving attacks on containerization tools, orchestration, and applications running in container systems threaten the successful implementation of such systems. Most contemporary attacks involve anomalous behavior in both individual processes and the whole system. The paper considers the problem of detecting anomalous behavior in container systems based on deep machine learning methods. It presents an approach for detecting anomalous process sequences within container systems using behavioral profiling techniques. This approach involves three key steps: analyzing frequencies of disassembled machine code instructions, building fixed-length process behavior histograms, and using these histograms for both training an Autoencoder – Long short-term memory hybrid model and performing detection tasks. The system generates histograms by counting disassembled instructions through frequency analysis, which calculates the ratio of specific instruction types to the total instruction count. After training on a profile reflecting normal container behavior, the neural network is ready to detect anomalous sequences of processes based on calculating the reconstruction error. The testing sequence of process histograms is fed into the neural network, which calculates the input vector's reconstruction error and compares it with a predefined threshold. The experimental results demonstrate both high detection accuracy and low computational resource requirements for neural network model training. The level of false positives is low, which allows using the proposed solution as an additional tool for detecting abnormal activity. The proposed approach allows for effective detection of attacks in which intruders intercept the execution of program code, change its behavior and manipulate function addresses in binary files.
Keywords: container systems, anomaly detection, process histograms, neural network, cybersecurity
Visitors: 1401

6. Combined method of deforming and animating objects based on positional modeling and continuum mechanics using free-form patches

[№3 за 2025 год]
Authors: Vyatkin, S.I. (sivser@mail.ru) - Institute of Automation and Electrometry, Siberian Branch, Russian Academy of Sciences, Synthesizing Visualization Systems Laboratory (Senior Researcher), Ph.D; Dolgovesov, B.S. (bsd@iae.nsk.su ) - Institute of Automation and Electrometry, Siberian Branch, Russian Academy of Sciences, Synthesizing Visualization Systems Laboratory (Head of the Laboratory), Ph.D;
Abstract: The paper presents a method for modeling the deformation and animation of objects based on positional modeling and continuum mechanics using free-form patches. The method is cross-functional for modeling diverse geometries and materials. The paper considers the problem of real-time deformation and animation modeling for rigid bodies, elastic materials, fabrics, and volumetric objects using constraint-based approaches. The subject of the research is the Euler and Newton methods, and how to use them in positional dynamics. The deformation and animation of three-dimensional objects are important functions, but they require a large number of calculations. The calculations employ a variational form of implicit Euler integration, processing constraints at the global level. This allows establishing connections between positional dynamics and the implicit Euler integration scheme. The local/global approach allows implementing implicit integration calculations without any special protection measures against singular or indefinite hessians to ensure the reliability of the method. This results in a simple implementation that does not need additional libraries to solve the problem and has little memory. Testing shows that after ten iterations, the modeling result visually looks similar to the convergent one using Newton's method. This enables using the method in real-time applications. Computer simulation demonstrate the method's practical effectiveness. Experimental results demonstrate that 5-10 iterations are enough for medium-scale model simulations, with each iteration requiring 1-6 milliseconds of computation time. The proposed method proves effective for deformation computation in solids, fabrics, shells, and similar materials. This is relevant when creating materials with the desired deformation behavior (animating clothes, characters and other soft forms). The proposed approach enables robotic tactile sensor modeling and MRI image simulation from anatomical models, which is crucial for validating image analysis algorithms.
Keywords: free-form patches, perturbation functions, computer modeling, positional dynamics, continuum mechanics
Visitors: 1188

7. Multi-agent simulation system for electric vehicle behavior modeling in heterogeneous distributed charging infrastructure in urban environment

[№3 за 2025 год]
Authors: Skorobogatchenko, D.A. (dmitryskor2004@gmail.com) - Volgograd State Technical University (Dr.Sci. (Engineering), Professor); Volchkov, S.N. (c4s23@yandex.ru) - Bionics and Neurotechnology LLC (Lead Developer); Safonova, E.V. (safonova_h@mail.ru) - Volgograd State Technical University (Lecturer);
Abstract: The paper investigates and applies simulation tools for multiagent modeling and computational analysis for electric autonomous vehicles. It examines the development of a simulation system to model interactions between a passenger electric vehicle fleet and a heterogeneous distributed public charging infrastructure in urban environment. The authors utilized data from a regional electric vehicle operator's hardware-software platform to acquire necessary datasets and construct models. The research considered various scenarios of urban electric vehicle traffic intensity, driver charging behavior, initial battery charge levels, and different configurations of external heterogeneous charging infrastructure. Additionally, the paper considers discrepancies in charging connectors between vehicles and charging stations, imposing extra constraints on driver charging behavior. The authors selected the SUMO simulation environment, implementing a program wrapper as an intermediate layer between SUMO and its standard TraCI Python library. The paper proposes an algorithm for rerouting an electric vehicle to a charging station based on its battery charge level and driver’s charging behavior determined during simulation. The authors developed a multiagent simulation model for charging stations, which differs from the traditional SUMO approach of using “charging lanes”. Instead, it models individual charging points at stations. This approach closely replicates the real-world process of charging electric vehicles using modern public charging infrastructure, where vehicles cease movement temporarily while occupying charging points. The proposed system enables the study of interaction patterns between autonomous electric vehicles and public charging infrastructure in urban agglomerations. This research holds significant practical value for developing urban autonomous electric transportation, addressing the interests of operators, city authorities and end-users of electric vehicles.
Keywords: multiagent simulation system, heterogeneous distributed electric vehicle charging infrastructure, mathematical and computer modeling, electric vehicle behavior model, software development, optimization of charging station placement
Visitors: 1372

8. Building an adaptive KOMPAS-3D CAD interface: A multilayer neural network

[№3 за 2025 год]
Authors: Subbotin, A.V. (aws1998@ya.ru) - Orenburg State University, Aerospace Institute (Postgraduate Student); Zubkova, T.M. (bars87@mail.ru) - Orenburg State University, Ph.D;
Abstract: The paper focuses on interface adaptation in CAD systems to enhance design efficiency for mechanical engineers in manufacturing environments. It presents a CAD interface adaptation approach for designer-specific tasks in KOMPAS-3D. The task is a set of data describing a future detail. The proposed approach's novelty is in employing a multilayer neural network to analyze relationships between input variables that describe part geometry and commands used during 3D model construction. Authors trained the neural network on 3D models from a standard parts library. The neural network generates a set of recommended commands that assist designers in creating part models. The developed software tool generates a neural network-driven toolbar interface that overlays the KOMPAS-3D CAD system window. A distinctive feature of this interface is the flexibility in configuration. The software tool enables expansion of the standard part database and retraining of the neural network to improve classification accuracy. The software tool integrates with KOMPAS-3D through its Software Development Kit (SDK) using Python. Deploying the developed software tool in mechanical engineers' workflows will optimize labor efficiency and boost productivity. The adaptive interface allows to reduce the number of commands and the time to search for them in the CAD interface. Thus, using a neural network to adapt the interface will increase the efficiency of interaction with KOMPAS-3D and provide a simpler and more intuitive workspace for a designer.
Keywords: multilayer neural network, interface, adaptive interface, design workspace, CAD system, Kompass-3D
Visitors: 1334

9. OpenFOAM with no graphical user interface: Solving the problem

[№3 за 2025 год]
Authors: Chitalov, D.I. (cdi9@yandex.ru) - South Ural Scientific Center (Junior Researcher);
Abstract: Since the standard OpenFOAM distribution kit has no graphical user interface, this fact complicates user interaction with the software. This paper aims to resolve this pressing issue. It presents an integrated module enabling numerical experimentation through the mulesQHDFoam solution framework. The developed solution offers two key competitive benefits: open-source code accessibility and Russian language interface. Separation of the program code that ensures frontend functioning from the backend program code simplifies subsequent support and refinement of the module. The paper describes some aspects of using mulesQHDFoam and peculiarities in designing experiments with this solver. It also presents a list of development tools including the primary programming language, supplementary libraries, and supporting technologies. A process diagram created using the UML diagramming language demonstrates the algorithm of the user's work with the program. The author tested the module's capabilities by simulating a fundamental continuum mechanics problem. The paper presents an image of the experimental results for the selected task with a visualization of the main window of the graphical shell. Since the source code is hosted on the GitHub, third-party specialists can use the program. The author shows the analysis of the effectiveness of the selected development technologies and determines their appropriateness for future implementation. He also presents concise results and deduces the implementation value of this research. He identifies several critical differentiators between developed modules and existing comparable software products.
Keywords: numerical simulation, openfoam, mulesQHDFoam solver, open source software, graphical user interface, python, continuum mechanics
Visitors: 1308

10. GitHub report classification using LogNNet reservoir neural network

[№3 за 2025 год]
Authors: Kovin, A.M. (akovin@list.ru) - Institute of Applied Mathematical Research, Karelian Research Centre of the RAS (Postgraduate Student); Ivashko, E.E. (ivashko@krc.karelia.ru) - Institute of Applied Mathematical Research, Karelian Research Centre of the RAS (Cand. of Sci. (Physics and Mathematics), Senior Researcher); Izotov, Yu.A. (izotov93@yandex.ru) - Institute of Physics and Technology, Petrozavodsk State University (Postgraduate Student, Research Associate); Velichko, A.A. (velichkogf@gmail.com) - Institute of Physics and Technology, Petrozavodsk State University (Cand. of Sci. (Physics and Mathematics), Associate Professor);
Abstract: Neural networks effectively handle classification tasks. However, their training and application require significant computing resources. Resource constraints particularly challenge embedded and wearable devices, as well as IoT edge computing systems. Natural language processing, especially text classification, is equally critical for edge computing. The paper analyzes the effectiveness of the LogNNet neural network using the example of classifying reports in the GitHub Issues task tracking system. LogNNet uses reservoir computing with auto-generated weights. The paper classifies reports into three categories: bug reports, improving issues, and software usage questions. The experiments involved using a 100,000-example database with imbalanced class distribution. The analysis established the minimum required feature dimensionality for adequate classification quality. Authors compared the results using accuracy, recall, f1-score, and perclass precision metrics against six standard machine learning methods: support vector machines, naive Bayes classifiers, k-nearest neighbors, decision trees, random forests, and logistic regression. The paper measures the RAM conservation enabled by LogNNet's compact memory representation. It occurred that LogNNet 100:50:20:3 reduces the amount of RAM used by 5 times. Meanwhile, the classification accuracy maintains 92% of the maximum values achieved by standard methods. LogNNet proves justified only for resource-constrained devices like microcontrollers and minicomputers when solving similar tasks.
Keywords: LogNNet, reservoir computing, machine learning, issue report classification, GitHub
Visitors: 1366

| 1 | 2 | Next → ►