Systems and Technologies

MOISTURE ESTIMATION IN SUGAR DRYING BASED ON A HYBRID MATHEMATICAL MODEL

G. V. Grygorchuk, L. I. Grygorchuk — Sat, 30 May 2026 00:00:00 +0300

The paper addresses the problem of moisture estimation in the sugar drying process, which is a critical stage of sugar production affecting product quality, storage stability, and energy efficiency. Direct continuous measurement of moisture in industrial conditions is difficult due to technological limitations and time delays associated with laboratory analysis. Therefore, the development of reliable indirect estimation methods is an important task for improving process control. A hybrid mathematical model for real-time moisture estimation is proposed. The model combines a physical description of heat and mass transfer dynamics with a neural network-based soft sensor used as a nonlinear correction element. The linear discrete model describes the main inertial behavior of the drying process, while the nonlinear component compensates for model uncertainties, parameter variations, and external disturbances. The structure of the model and the data processing algorithm are presented. Simulation studies were performed taking into account measurement delays typical for industrial conditions. The influence of the discretization step on estimation accuracy and control performance was also analyzed. The results demonstrate that the hybrid model reduces the moisture estimation error by approximately 25–30 % compared to the linear model and provides stable performance under varying operating conditions. The proposed approach enables reliable estimation of an unmeasured state variable and can be integrated into industrial control systems. The developed model is suitable for real-time applications and can be used for monitoring, stabilization, and optimization of the sugar drying process.

BILATERAL ESTIMATES OF THE MAXIMUM LYAPUNOV INDICATOR

S. Yu. Poslavskyi, D. O. Redchyts, O. V. Akimenko, S. V. Moiseienko — Sat, 30 May 2026 00:00:00 +0300

Many systems encountered in problems of mechanics, control theory, and other fields of science and engineering are described by nonlinear differential equations with time delay. Most studies consider systems with constant delay; however, information about the delay function is often unavailable, with only its upper bound known; furthermore, the system may contain distributed delay. Known methods, in most cases, allow one to obtain only sufficient conditions for stability or upper bounds on the maximum Lyapunov exponent. A drawback of such results is that the degree of their conservativeness remains unknown. In this regard, the task of localizing the maximum Lyapunov exponent is relevant; that is, in addition to an upper bound, it is necessary to find its lower bound. The proximity of these bounds guarantees the accuracy of the obtained upper bound and, consequently, of the sufficient stability conditions. The article provides a detailed examination of a class of systems of nonlinear differential equations with a given linear part and a norm-bounded nonlinear term containing a variable delay. Particular attention is paid to the effect of the delay on the system’s dynamics and the estimation of its stability characteristics. Was obtained two-sided estimates of the maximum Lyapunov exponent, expressed in terms of the norm of the nonlinear term as well as in terms of the maximum values of the delay functions. This allows us to establish quantitative bounds on the behavior of the solutions and to estimate the rate of their convergence or divergence. For certain classes of systems, it was possible to determine exact values of the maximum Lyapunov exponent, which is an important result for stability theory. Based on the obtained estimates, sufficient, and in certain cases necessary, conditions for the exponential stability of the studied systems have been formulated. A characteristic feature of these conditions is their invariance with respect to delay, which significantly expands their scope of application. A simple and effective method for verifying exponential stability is also proposed, which does not require complex calculations and has a computational complexity that is practically independent of the system’s dimension (order). This makes the approach convenient for practical use, particularly for high-dimensional systems. Finally, a series of examples is provided that illustrate the application of the developed method, demonstrate its effectiveness, and confirm the theoretical results.

AUTOMATED WORK WITH FINITE AUTOMATA: A TOOL FOR TEACHING IN THE ERA OF ARTIFICIAL INTELLIGENCE

T. M. Sopronyuk — Sat, 30 May 2026 00:00:00 +0300

The article addresses the problem of maintaining objectivity in student assessment in the context of widespread access to artificial intelligence tools, particularly large language models such as ChatGPT. The study focuses on the discipline of formal languages and automata theory, where traditional task formats (transition tables) are easily solved by modern AI models. The purpose of the work is to develop software for automated generation, determinization, and visualization of finite automata, enabling rapid creation of individual assignment variants in visual format (transition diagrams). The system includes a random NFA generator with guaranteed state reachability and controlled transition density, a visualization module with customizable parameters, a pairwise automaton display function for concatenation and union operations, an assignment variant generator (20 variants with concatenation, union, and iteration tasks), and an NFA determinization algorithm that reads automata from text files and visualizes the resulting DFA. Experimental results show that visual format tasks significantly reduce AI solution success, while student testing confirms reduced AI usage. The developed software can be used in teaching courses on automata theory and language processors.

ON THE METHODOLOGY FOR CALCULATING THE AERODYNAMIC CHARACTERISTICS OF AN UNMANNED AERIAL VEHICLE IN THE EARLY STAGES OF DESIGN

A. V. Sokhatskyi — Sat, 30 May 2026 00:00:00 +0300

Determining the aerodynamic characteristics of unmanned aerial vehicles is an extremely complex problem. It is believed that the most accurate way to determine the aerodynamic characteristics of any vehicle is through computational modeling. This procedure is performed on high-performance computers using Reynolds-averaged Navier–Stokes equations. However, this process is complex and costly. It requires significant time for developing methodologies, constructing algorithms, writing software packages, and the actual operation of the computers. In the numerical solution of aerodynamic problems using the Reynolds-averaged Navier–Stokes equations, it is not the differential equations themselves that are solved, but their finite-difference analogs. It is necessary to correctly enforce the physical conservation laws of mass, momentum, and energy. In addition, a specific turbulence model must be applied. This is a challenging task for mesh-based methods for solving problems in mathematical physics. Designing the aerodynamic configuration of an unmanned aerial vehicle requires solving an inverse problem. Currently, no such methods exist. Problems involving the search for the optimal aerodynamic configuration of a transport vehicle for any purpose are solved iteratively through a process of gradual approximation. This requires significant time and financial resources. To calculate the aerodynamic characteristics of unmanned aerial vehicles in the early stages of design, it is proposed to use empirical approaches. This paper presents a methodology for calculating the aerodynamic characteristics of an unmanned aerial vehicle. To develop this methodology, algebraic relationships obtained from experimental studies of aircraft aerodynamics by domestic and foreign researchers were utilized. To perform the necessary calculations, algorithms were developed and a software package was written in the Fortran-95 programming language. Calculations of the aerodynamic characteristics of the fuselage for an unmanned aerial vehicle were performed and compared with experimental data.

NONLINEAR ALGORITHMS FOR LOAD DISTRIBUTION TO WIND TURBINE STABILIZATION CHANNELS

S. V. Tarasov, O. N. Molotkov — Sat, 30 May 2026 00:00:00 +0300

Today, wind power plants are increasingly taking up a place among operating renewable energy systems. To achieve their most successful operation, it is extremely important to use the maximum available wind power to achieve wind turbine (WT) operation at maximum power. Algorithms for tracking the maximum power point and algorithms for stabilizing the operation of the WT in a small vicinity of this point are extremely important in this context. To ensure small deviations in the rotation speed around the maximum power point for the current wind flow speed, the traditional means is to use a generator to control the rotation of the WT rotor. The use of changes in the turbine rotor configuration for this purpose is a fairly new direction in the design of control systems. Researchers note a number of useful properties of WTs with a variable rotor configuration, which make them quite promising for further use. The difficulties of designing such VTs are associated with the increased methodological complexity of the analysis and synthesis of rotor rotation control systems – general mathematical models of the rotor are nonlinear, which is associated with the dependence of the rotor inertia tensor on changes in generalized coordinates, and their linearization leads to non-stationary models. Unfortunately, there is no comprehensive description of such systems. A possible way to reduce the complexity of the description is to reduce the maximum changes in the length of the traverses while ensuring the loading of both stabilization channels – the channels for changing the length of the blades and the traverses during their joint operation. The purpose of the article is to analyze the stability of dynamic algorithms and quality indicators of the regulation system for stabilizing the rotation of the Darie VT rotor of variable configuration, as well as the conditions for the absence of a static regulation error. The methods for solving the problem are methods of the classical theory of automatic control and mathematical modeling. The novelty of the obtained results lies in the constructed dynamic stabilization algorithms, the conditions of their stability and the absence of a static regulation error, as well as in the dissemination of the method of load distribution on stabilization channels with algorithms that dynamically The conducted mathematical modeling proved a significant improvement in the dynamic properties of the variable configuration Darier rotor rotation stabilization system using dynamically changing algorithms.

COMPUTER MODELING OF EPIDEMIC SPREAD BASED ON CELLULAR AUTOMATA

Y. V. Ulianovska, T. M. Rudianova, E. A. Riabovolenko, V. O. Striukovatskyi — Sat, 30 May 2026 00:00:00 +0300

The paper provides a comprehensive analysis of modern methodological approaches to the computer modeling of infectious disease spread processes, with the mathematical apparatus of cellular automata serving as the pivotal tool. The author substantiates the scientific expediency of transitioning from classical analytical models of the SIR type, characterized by a certain level of abstraction and simplification, to more flexible discrete-spatial models. Such models allow for a significantly more adequate reproduction of the complex and non-linear spatio-temporal dynamics of epidemic processes, taking into account the structural heterogeneity of the environment and the stochastic nature of interpersonal contacts within a population. Within the framework of the conducted research, a comparative review of existing scientific paradigms was carried out, specifically multi-agent modeling concepts, which enabled a clear definition of the advantages of the cellular automata approach in tasks involving the mapping of local interactions, diffusion processes, and the direct impact of physical spatial constraints on pathogen transmission. The primary scientific result of the work is the development and full-featured software implementation of an interactive model of infection spread, which provides high-quality visualization of the epidemic process dynamics in real-time, supporting the functionality for operational changes in simulation parameters. The model architecture is based on a two-dimensional state matrix structure, where each individual cell is identified as an autonomous agent and can reside in one of the predefined epidemic states: susceptible, infected, immune (recovered/removed), deceased, or a stationary barrier. Transition rules between states are formalized based on developed probabilistic mechanisms of infection transmission and temporal characteristics of the disease course, ensuring high flexibility in adapting the model to various scenarios for both viral and bacterial outbreaks. The software implementation of the model was performed using the Python programming language, employing specialized NumPy libraries for optimizing matrix calculations and Pygame for implementing the graphical user interface and interactive visualization. During the study, a series of complex computational experiments were conducted, including varying the initial vaccination levels, changing the intensity of social contacts, and modeling the implementation of quarantine restrictions of various degrees of severity. The obtained empirical results confirm the adequacy of the developed model, particularly its ability to accurately reproduce characteristic wave-like epidemic dynamics, outbreak localization effects, and the formation of herd immunity. It has been established that the synergetic combination of high preventive immunization levels and timely restrictive measures is the most effective factor in containing an epidemic threat. The practical significance of the developed software complex lies in the possibility of its wide application as a tool for fundamental scientific research, short-term forecasting, and for educational purposes for specialists in relevant profiles. Prospects for further investigation in this direction are associated with the deep integration of the cellular automata model with modern machine learning methods and Big Data analysis to increase predictive accuracy based on real statistical indicators of urban mobility.

MODELING THE TRANSFORMATION OF PROGRAM PERSONNEL MANAGEMENT PROCESSES IN A MULTIPROJECT ENVIRONMENT

N. V. Dotsenko, Y. S. Lutsiv — Sat, 30 May 2026 00:00:00 +0300

The article is devoted to the study of the transformation of HR management processes of programs in a multi-project environment. The purpose of the study is to develop a model of the transformation of HR management processes of programs in a multi- project environment. The study is based on the use of the project and program approaches to managing program projects, the methodological support for business process reengineering, the combinatorial analysis and optimization methods. The scientific novelty is the development of a set of models of transformations of management processes: a contextual model and a decomposition model of the transformation of HR management processes of programs in a multi-project environment. The study yielded results that contribute to improving the effectiveness of transformation. An analysis of approaches to studying transformation processes identified aspects that should be considered when developing an adaptive digital HR ecosystem. The structure and functions of an adaptive digital HR architecture are defined. Typical characteristics of an adaptive digital HR architecture for a multi-project environment are identified, and their classification is proposed. To formalize transformations, the contextual model for transforming HR management processes for programs in a multi-project environment was developed. A decomposition of the transformation processes was performed, and a model for transforming HR management processes for programs in a multi-project environment was constructed. Elements of a multi-agent transformation system were identified, and information support for transformation was considered. The application of reengineering to manage transformation processes is considered. The main stages of transformation are identified. Using the proposed models will formalize transformation processes and identify the starting point for transformations. The created TO BE process models, a register of transformation changes, and transformation recommendations will provide information and methodological support for management decision-making.

MODELING AND OPTIMIZATION OF DYNAMIC CONTROL IN STOCHASTIC SYSTEMS BASED ON MACHINE LEARNING

S. I. Zhyr, G. A. Shyshkanova, T. A. Zaytseva, T. I. Sliusarova — Sat, 30 May 2026 00:00:00 +0300

The problem of optimizing the functioning of the entry group of a container terminal operating under conditions of stochastic uncertainty of transport flows caused by global logistics trends and random external factors is considered. The relevance of the chosen direction is due to the need to reduce truck waiting time, minimize operating costs and level the negative environmental impact from excess emissions during transport downtime in queues. Obviously, a simple expansion of the physical infrastructure is often economically impractical. A transition to an intelligent hybrid dynamic control model based on the reinforcement learning (RL) paradigm is proposed, which allows the system to adaptively regulate the number of active service channels. The developed model is based on a Markov decision-making process. To adequately reproduce the real dynamics of truck arrivals, a Poisson distribution is used. The entry group is represented through a discrete approximation of the classical mass service model. The use of Q-learning ensures finding the optimal control policy even in the absence of exhaustive a priori information, allowing the agent to «learn» directly during interaction with the environment. The study illustrates the evolution of agent learning and confirms its convergence to a theoretically justified optimal strategy. The modeling results show that the implementation of RL methods contributes to effective smoothing of peak loads, a significant reduction in queue length and an overall increase in terminal throughput. The possibilities of scaling the model through the integration of deep neural networks are considered, which allows operating with large data sets and complex state spaces. The Hamilton–Jacobi–Bellman equation, which determines the optimality limits in continuous control problems, is a theoretical verification of the obtained strategies. The proposed approach has practical significance for the development of logistics systems, as it allows integrating hybrid intelligent algorithms into infrastructure management and ensures optimization of economic indicators.

PROCEDURAL AND GENERATIVE METHODS FOR GAME CONTENT CREATION

H. A. Zavhorodnia, Ya. I. Kornaga, V. V. Zavhorodnii — Sat, 30 May 2026 00:00:00 +0300

This paper investigates contemporary methods of procedural and generative game content creation, with a focus on rulebased approaches and stochastic models, as well as the challenges of integrating these methods to improve generation efficiency. Special attention is given to the problems of scalable content generation under limited computational resources and high requirements for variability, structural consistency, and realism of game environments. The study analyzes current trends in automatic content generation, including algorithmic rules, probabilistic models, Markov processes, stochastic grammars, and hybrid approaches that combine the advantages of deterministic and random mechanisms. The work proposes a formalization of the content generation process as a composition of deterministic and stochastic operators, allowing for simultaneous improvement of controllability and diversity of results. A mathematical model of the generative process is introduced, based on a state distribution function and a constraint function, ensuring the validity of generated game configurations. The study demonstrates how combining rule-based systems with stochastic generators enables a balance between predictability and variability of results, providing flexible adaptation to user requirements and dynamic game processes. Particular attention is paid to analyzing the efficiency of different approaches in terms of computational complexity, scalability, and quality of generated content. A series of experimental studies were conducted comparing the proposed hybrid model with baseline generation algorithms, including pure rule-based systems and random stochastic generators. The results show that hybrid methods can significantly increase content diversity (by 35–50 %) while maintaining structural integrity, and also reduce generation time compared to classical approaches. The study confirms that integrating deterministic and stochastic mechanisms is an effective approach to improving the quality of procedural content in modern game systems. The obtained results can be applied in the development of procedural worlds, adaptive generative environments, and modern computer games, ensuring a balance between predictability, variability, and computational efficiency.

STOCHASTIC OPERATORS OF ACTION IMPACT: FORMALISATION AND MULTI-STEP REGULARISED LEARNING

B. Yu. Zaika, S. V. Yershov — Sat, 30 May 2026 00:00:00 +0300

This paper proposes a formalisation of action impact in stochastic dynamical systems as a dedicated stochastic operator acting on system states. Accurate modelling of action impact is an important problem in sequential decision-making under uncertainty, since in many real-world systems actions are applied repeatedly and their consequences propagate through system dynamics over time. While modern machine learning approaches, including reinforcement learning and conditional density estimation, can approximate short-term transitions, the behaviour of learned models under recursive multi-step application remains insufficiently studied. In most existing frameworks, transition dynamics are embedded within policy optimisation or trajectory prediction objectives and are rarely treated as independent modelling entities. In the proposed approach, the action impact operator maps the current system state and applied action to a conditional distribution of future states and is defined with explicit compositional structure. This enables the analysis of recursive operator application across multiple time steps. A learning objective is introduced that combines one-step negative log-likelihood with a multi-step consistency term derived from operator composition. The central hypothesis of the study is that one-step maximum likelihood training does not guarantee stable long-horizon behaviour when the learned operator is recursively applied. To investigate this hypothesis, empirical evaluation is conducted in a fully observable stochastic dynamical system using a minimal realisable linear Gaussian model. The empirical results show that purely one-step training leads to long-horizon degradation, including accumulation of trajectory error and systematic underestimation of predictive uncertainty. Introducing explicit multi-step regularisation significantly improves long-horizon stability and uncertainty calibration, and the improvement persists beyond the training horizon. The proposed formulation establishes a basis for modelling action impact in stochastic dynamical systems and provides a machine-learning framework for robust modelling of recursively applied transitions. This provides a foundation for further research in partially observable environments, nonlinear architectures, and decision-support systems.

A SET-THEORETIC APPROACH TO MODELING CASCADING DERIVATIVE RISKS IN SOCIO-TECHNICAL SYSTEMS

V. I. Ziuziun, D. D. Shcherbak — Sat, 30 May 2026 00:00:00 +0300

The article addresses a pressing scientific and applied problem: modeling complex hierarchical relationships among risk factors arising in the operation of high-tech, socially oriented systems. The object of the study is a network of sorting stations, considered as a complex dynamic system with distributed business processes. The relevance of the research is driven by the high level of turbulence in the external environment and the need to move from qualitative descriptions of risks to their digital formalization and quantitative measurement. The proposed approach is grounded in systems analysis, which enabled a multi-level decomposition of the organizational structure into subsystems of financial planning, logistics, sales, and marketing. Set theory was employed as the mathematical framework to describe the interactions among these subsystems. This made it possible to represent the cascading development of risks as a sequence of system states in which the emergence of a primary threat (funding shortfall) initiates a set of derivative risks, ranging from technological degradation of sorting lines to the loss of intellectual capital. The scientific novelty of the work lies in the further development of set-theoretic membership models that establish logical relationships between the causes and consequences of critical situations in a format suitable for automated processing. For the first time, an algorithm for the quantitative assessment of cascading impacts has been proposed through an integral indicator of Expected Risk Value (ERV), based on a combination of probabilistic characteristics and degrees of influence on the system’s target performance indicators. The practical significance of the research is realized in the form of a strategic management map-scheme representing a set of algorithmized response strategies (avoidance, mitigation, acceptance). The proposed measures integrate both managerial decisions and technical-technological innovations, including the use of robotic systems to automate sorting processes. This reduces critical dependence on the human factor and minimizes operational risks. The application of the developed models creates a mathematical foundation for the design of intelligent Decision Support Systems (DSS) capable of predictive monitoring of complex systems and the automatic generation of cascading threat neutralization scenarios. The use of such digital tools ensures not only the short-term stabilization of business entities but also creates conditions for their sustainable development and enhanced competitiveness in the context of the economy’s digital transformation.

ALGORITHMS AND METHODS OF ADAPTIVE BEHAVIOR OF NON-PLAYER CHARACTERS IN COMPUTER GAMES

Y. Y. Iliash, V. А. Rovinsky, A. O. Heiko — Sat, 30 May 2026 00:00:00 +0300

The article is devoted to a comprehensive study of the Behavior Tree model as a tool for implementing artificial intelligence of Non-Player Characters (NPCs; Agents) in modern computer games. The relevance of applying hierarchical decision-making architectures is substantiated in the context of increasing demands for interactivity, adaptability, and realism in NPC behavior. It is demonstrated that the structured behavior tree model ensures logical organization of an agent’s actions, a transparent prioritization mechanism, and the possibility of system scalability without loss of manageability. The paper proposes a formalized architecture of a behavior tree, distinguishing composite nodes such as Selector and Sequence, as well as leaf nodes that implement specific actions (pursuit, attack, patrol, idle/wait). The tri-state execution model (Success, Failure, Running) and the tick-based evaluation principle are analyzed in detail, ensuring real-time system reactivity. Particular attention is given to constructing a priority hierarchy of behaviors, in which combat scenarios have higher significance, while alternative states serve as background or fallback actions. The practical implementation of the model was carried out in the Unity environment using the C# programming language. The uniqueness of the proposed algorithm implementations lies in the adaptability of the code for other engines based on C++. Examples of programmatic implementation of the base node class, composite structures, and tree initialization are presented, demonstrating the correspondence between theoretical principles and real software implementation. It is shown that the proposed architecture is modular, extensible, and suitable for integration into game projects of varying complexity. The obtained results confirm that behavior trees allow combining algorithmic rigor with game design flexibility, ensuring predictable yet dynamic agent behavior. The proposed approach can be used in educational prototypes, indie projects, and commercial developments, and can also serve as a foundation for further research related to Behavior Trees.

SOFTWARE IMPLEMENTATION OF A DESKTOP INFORMATION SYSTEM FOR LIBRARY AUTOMATION

V. S. Morokhovych, D.-I. V. Pavlyk — Sat, 30 May 2026 00:00:00 +0300

This article describes the development of a desktop solution for the automation of library operations, which optimises the management of the collection, users, book circulation and staff work. The system provides for the authentication of users with the roles of librarian and administrator, with data verification via a secure connection to the database. The application implements functionality for viewing, adding, editing and searching for information about readers, books and loan transactions, with access to certain actions restricted depending on the user’s role. The “Loans” tab supports automatic calculation of service charges based on the duration of book use, determination of book return or loss status, and the imposition of fines for late returns. The administrative section of the system allows you to manage library staff, change passwords, dismiss employees, and update reference information on genres and publishers. The database structure is also described, error handling is implemented, and a reliable connection to the database is ensured. Thanks to its intuitive interface, clear separation of roles, and extensive functionality, the information system significantly improves the efficiency of library operations, ensuring transparency and convenience in day-to-day tasks. The technical methods described can be adapted for use in related fields, such as archival record-keeping, the development of local CRM systems or electronic card catalogues, where the key requirements are autonomy, performance and data security. The primary focus of the application’s development is to expand its functionality whilst maintaining full autonomy and stable operation within a local network. This is particularly important for libraries that do not have constant access to the internet or operate in environments with stringent requirements for data storage on internal media.

VISUALIZATION OF THE ALGORITHM FOR MUNICIPAL GREEN BOND ISSUANCE IN UKRAINE: A GRAPH-AS-A-CODE APPROACH

T. V. Ratushnyak, О. V. Hladchenko, N. R. Golovko — Sat, 30 May 2026 00:00:00 +0300

In Ukraine, despite the legislative implementation of green bonds since 2020, the market remains in its infancy, represented only by isolated corporate cases. There is a critical gap between international trends in scaling sustainable investment and the domestic financial environment, which is constrained by systemic barriers. Ukrainian territorial communities, as potential public issuers, lack effective algorithms for issuing municipal green bonds. In the context of post-war reconstruction based on environmental modernization, this financial instrument acquires strategic importance for attracting capital into renewable energy, energy efficiency, and waste management. However, the high complexity of the emission procedure creates significant risks of management errors, highlighting an urgent need for the visualization of financial and legal processes. The purpose of the study is the scientific substantiation and development of a methodology for the dynamic visualization of the municipal green bond emission algorithm in Ukraine based on the «Graph-as-a-Code» concept. This involves using generative artificial intelligence to transform semantic legal descriptions into declarative Mermaid code to minimize management risks and overcome institutional barriers. The study employs a multidisciplinary approach based on systemic analysis and logical generalization to identify barriers in the sustainable finance market. Structural-functional modeling was used to decompose the emission algorithm into six stages in accordance with budget and financial legislation. The practical implementation of the model is based on semantic analysis combined with prompt engineering, ensuring the transformation of legal descriptors into valid Mermaid code. The final stage involves graphical visualization and rendering in digital environments to convert large arrays of text into interactive diagrams. The integration of the Gemini multimodal AI model allowed for the intellectual convergence of multi-source data, eliminating information gaps. The study proposes a technology for visualizing the emission algorithm through the «Graph-as-a-Code» method, implemented via five operations: from forming the semantic description to automatic rendering on GitHub. The research resulted in a dynamic visualization of the algorithm, which allows municipalities to perceive the emission process as a clear path rather than a wall of restrictions. The scientific substantiation and practical testing of the dynamic visualization methodology prove that traditional modeling methods (UML, BPMN) require adaptation to modern digital environments through declarative markup languages. The use of AI for code generation ensures high accuracy in reflecting complex financial and legal processes. The proposed approach minimizes management risks for local self-government bodies and ensures transparency in emissions.

RELIABILITY-DRIVEN STREAMING MODELLING OF MULTIMODAL TIME SERIES FOR ROBUST DECISION SUPPORT UNDER DRIFT AND MODALITY DEGRADATION

I. S. Uzun, M. V. Lobachev — Sat, 30 May 2026 00:00:00 +0300

Streaming decision support systems that process multimodal time series must remain robust under simultaneous concept drift and temporary modality degradation. Existing approaches usually treat multimodal fusion, drift adaptation, anomaly detection, and probability calibration as separate problems, which makes it difficult to distinguish whether a loss of predictive quality is caused by a real change in the process or by a temporary failure of one modality. This paper presents a unified online pipeline in which online reliability estimation is used as a single control interface for reliability-adaptive dynamic fusion, driftinitiated budgeted micro-adaptation, and reliability-constrained anomaly detection under a false alarm rate budget. Reliability is modeled as a causal probability of the current non-degraded modality state, post-hoc calibrated, and reused in all downstream control rules. Evaluation follows the prequential protocol on controlled streams with deterministic injections of modality degradation, concept drift, and anomalies, and on real data from UCI Appliances Energy Prediction and UCI Air Quality. The calibrated reliability model retains high degradation-separation ability (ROC - AUC = 0.8624) and improves calibration to ECE = 0.0845 versus 0.1840 without calibration. RADF preserves clean-regime quality (MAE = 0.5557; RMSE = 0.6952) and improves degraded segments, for example MAE = 0.6402 versus 0.6820 for early fusion under alternating missingness. Budgeted micro-adaptation improves post-drift forecasting relative to no adaptation (MAE = 0.6613 versus 0.7046; average recovery 157 versus 800 steps) while updating only a three-parameter head within fixed budgets. RC-AD increases Recall@FAR at all tested budgets, including 0.335 versus 0.103 at a false alarm rate budget of 0.05 on controlled streams. In the integrated stress scenario, the system achieves ROC - AUC = 0.983, ECE = 0.029, Recall@FAR=0.311, and PR - AUC = 0.287; in robust application protocols with severe segment missingness, it reduces MAE by about 35 % in the energy domain and about 93 % in the BTC/USD domain relative to naive fusion. These results present one coherent streaming decision support result rather than isolated local methods.

THE APPLICATION OF INSTANCEDMESH FOR VISUALIZING THE DYNAMICS OF PROCESSES IN THREE-DIMENSIONAL CELLULAR STRUCTURES

O. D. Firsov — Sat, 30 May 2026 00:00:00 +0300

In the present study, an algorithm along with its software implementation has been developed for the visualization of specialized three-dimensional cellular structures. The solution utilizes the capabilities of the INSTANCEDMESH class provided by the Three.js library in the JavaScript programming language. The object of research is the process of constructing models of cellular structures that evolve in accordance with theoretical algorithms, resulting in corresponding changes to their visual representation. Within the framework of this work, an algorithm for visualizing a three-dimensional array has been proposed, implemented, and tested. In this array, cell values are updated according to specific time steps (tacts) of the modeled system’s operation. The research findings demonstrate that array processing technology extended by the INSTANCEDMESH class serves as an effective tool for the three-dimensional visualization of cellular structures of limited size. It enables the creation of dynamic visualizations that reflect changes occurring during the modeling process. The issue of visualized structure size has been analyzed in detail. In the three-dimensional case, the volume of data required to identify significant structures grows according to a third-degree polynomial, in contrast to standard one-dimensional cellular automata. While this growth facilitates the organization of the modeling process, it simultaneously introduces challenges in visualizing the dynamics of changes, particularly regarding the ability of a human analyst to interpret the resulting three-dimensional image effectively. The initial number of cells that permits meaningful analysis of the generated constructions ranges from 10⁴ to 10⁶. The visualization of cellular structures is constrained not only by computational performance but also by the necessity to render each individual element with sufficient clarity, ensuring a distinct perception of both separate elements and the overall construction. The proposed solution can be applied to the modeling of dynamic processes that are amenable to formalization and representation as voxel-based structures. Approaches to the organization of data processing and visualization for three-dimensional cellular structures – taking into account both performance requirements and the accessibility of each representational element for analysis – have been successfully implemented and tested. The obtained results will support further computational experiments aimed at analyzing the development of three-dimensional cellular automata and analogous structures.

AN INTERACTIVE SYSTEM FOR SEMANTIC EDITING OF RASTER GRAPHICS BASED ON THE INTEGRATION OF MULTIMODAL GENERATIVE APIS

D. V. Chornobryvets, S. V. Popereshnyak — Sat, 30 May 2026 00:00:00 +0300

This paper investigates the process of automated semantic editing of raster images using artificial intelligence methods. The relevance of the study is обусловлена high computational demands of modern generative models, which require powerful graphics processing units (GPUs) to perform inpainting operations, as well as the limited flexibility of cloud-based services in terms of precise spatial control over editing. This creates a significant barrier to the accessibility of intelligent image editing tools for users with limited computational resources. The aim of the work is to improve the accessibility and efficiency of semantic image editing processes by distributing the computational workload between the client-side and server-side components of the system. The proposed approach combines local tools for spatial mask generation with the use of a cloud-based multimodal API (Gemini 3 Flash Image) to perform generative transformations. As a result, a lightweight desktop application with a modular client-server architecture has been designed and implemented. The key features of the system include asynchronous multithreaded processing of network requests, which ensures a responsive graphical user interface, as well as the use of reverse compositing algorithms for seamless integration of generated fragments into the original image. A real-time binary mask generation mechanism based on cursor coordinates has been implemented, enabling high-precision selection of regions of interest. The obtained results are explained by the effective offloading of tensor computations to cloud infrastructure while maintaining local control over the editing process. Experimental evaluation confirmed the feasibility of performing complex image transformations on low-performance devices without loss of output quality. The practical significance of the work lies in the possibility of using the developed system by digital artists, designers, and researchers for rapid prototyping and image editing without the need for specialized hardware.