The natural classification of scientific knowledge has always been a challenging task. Its resolution would enable the organization of existing knowledge and facilitate the extraction of new insights for research, practical application, and education. This is particularly important in economics, which is seeing a proliferation of new fields that study national economic development using modern methods of systems analysis and mathematical modeling. This paper proposes a triadic classification scheme based on the principle of fractal similarity, which allows the entire structure to be replicated at every taxonomic level. Within this scheme, only typological units are treated as disjoint layers of knowledge; all other knowledge groups constitute combinations or combinatorial associations of these independent layers. To articulate this structure, we employ stratification procedures from differential geometry, separating information into distinct layers with different contents. The hierarchical classification comprises several levels, progressing from simple to complex and from abstract to concrete: data, concepts, models, intertheories, metatheories, and metascience (exemplified by mathematics). The metatheory level is divided into three sectors – general, quantitative, and formal knowledge – which correspond to political economy, financial analysis, and mathematical (digital) analysis, respectively. These sectors account for the economy's linkages with the external and internal environments of business entities. In a cross-cutting manner, intertheories describe phenomena in both nature and society as specific types of systems using a uniform terminology. On this basis, models of various economic entities – including enterprises in the energy sector – can be constructed. The data and knowledge representation space are defined by eight independent coordinates, one of which is economic growth. This framework underscores the independent meaning of economic knowledge, distinguishing it from that of other sciences.

Survival in a changing environment is a task that requires identifying and remembering the most significant regularities of the environment and acting with them in mind. The ability of an organism (or an agent imitating an organism) to form and use internal representations of the external environment is called reflection in a broad sense. The article is devoted to identifying the connection between the predictability of events in the environment and the emergence of a universal skill of reflection in agents in this environment. As agents, we used heuristic model objects – simple recurrent neural networks, the primary training of which was carried out within the limit world models. These models were implemented as a set of tasks (in order of increasing predictability): responding to a random sequence of stimuli, reflexive game, responding to a set of fixed sequences, and responding to a single fixed sequence. The emergence of the universal skill of reflection after the primary training on each of these tasks was assessed by the success of the trained model objects in test tasks. The maximum Jacobian eigenvalue of the weight matrix and the type of the neural activity trajectory after the initiating single stimulus were regarded too. Based on this set of characteristics, we found out that world models with a predetermined periodicity of events (stimuli) contribute to the greatest extent to the formation of the universal reflection skill. On the contrary, in unpredictable environment conditions the emergence of internal representations is baffling. These results lead to the better understanding of reflection in a broad sense and simplify the choice of environmental conditions for further research of this phenomenon.

The relevance of this study is determined by the fact that the efficiency of classical genetic algorithms (GAs) in solving global optimization problems significantly depends on the choice of crossover and mutation probabilities, while fixed parameter values often lead to premature convergence and stabilization of the population in the vicinity of local extrema. The aim of this work is to develop and experimentally evaluate approaches to adaptive control of GA parameters based on artificial neural networks and reinforcement learning. Within a unified mathematical formulation, population state features, a set of actions represented by discrete changes in mutation and crossover probabilities p_m and p_c, and a reward function reflecting the improvement in solution quality between generations with a penalty for excessively high mutation are defined. Four algorithmic variants are considered: a classical GA with fixed parameters, a hybrid GA with a neural-network controller (GA+NN), a GA with tabular Q-learning (GA+RL), and the proposed parameter control method based on deep Q-learning, which uses a neural network to approximate the Q-function (GA+DQN). The scientific novelty of the study lies in the integration of a DQN agent into the GA parameter control loop within a formalized “state–action–reward” model and in comparing its efficiency with that of a neural-network controller and tabular Q-learning on continuous optimization problems. Numerical experiments were carried out on the Rastrigin and Schaffer test functions with 20 independent runs for each configuration. The final metrics used were the best objective function value in the last generation f_min(T_max) and the best value achieved over the entire run. It is shown that GA+RL provides the greatest improvement in solution quality. The GA+DQN method demonstrates a moderate improvement over the baseline GA, confirming the applicability of deep Q-function approximation to parameter control. The neural-network controller in the considered training scheme shows high sensitivity to parameter settings and, in these experiments, performs worse than the reinforcement learning approaches. The comparison results are presented in the form of convergence plots, an analysis of population diversity indicators, and a summary table.

One of the tasks associated with developing an interactive collaborative robotic manipulator is temporal analysis of the work scene, i.e., determining the order in which objects appear (and disappear) within the field of view of the vision system. Traditionally, this issue has been considered in the context of satellite imagery and has not been sufficiently addressed in the literature with regard to scenes located approximately 1 m from the camera. At the same time, work scene analysis based on bi-temporal images is a relevant area of research in the context of the development of robotics in general and physical artificial intelligence in particular. Creating high-quality temporal change masks of the work scene makes it possible to determine the contours and geometric centers of new objects for subsequent grasping by the robotic manipulator. A high-quality temporal mask should not contain falsely detected change regions (change objects that do not actually exist), yet should clearly outline the contours of genuine change objects in the work scene. The paper presents a mathematical formulation of the temporal analysis problem and, on its basis, proposes a method for generating temporal change region masks by differentiating “before” and “after” images, combining classical computer vision techniques with the neural network segmentation model SAM (Segment Anything Model). The novelty of the proposed approach lies in applying to the difference image not algebraic processing, but rather its segmentation into two regions (a change region and a no-change region) using a neural network segmentation model. The proposed approach was compared with algebraic methods for creating temporal masks (Change Vector Analysis – CVA and Slow Feature Analysis – SFA) and with the use of a multilayer perceptron neural network architecture (input layer of 12 neurons, hidden layer of 512 neurons, output layer of 1 neuron). It is demonstrated that the proposed approach enables the generation of high-quality change masks for diverse objects against a large number of backgrounds (including cluttered ones), a result that is difficult to achieve with the methods brought for comparison. At the same time, the proposed approach can be implemented “on the fly,” i.e., in real time during robot operator work, only if a Graphics Processing Unit (GPU) is available.

This paper aims to describe the development and training of a model capable of classifying vehicles by make and model based on images, as well as an interface for convenient interaction with the model. The novelty of the work is demonstrated by the use of modern vehicle identification methods for fine-grained classification. The paper is divided into three sections, each covering key aspects of the research. The first section provides a domain analysis, reviewing existing vehicle identification methods, including lidar-based technologies, as well as methods using images and videos. Particular attention is paid to the analysis of modern approaches to classifying vehicles by make, model, and other attributes. This analysis allowed us to identify the strengths and weaknesses of various approaches and justify the choice of deep learning architectures for further research. The second section describes the collected dataset with detailed labeling by make and model used for the study. Three machine learning models are compared using various metrics, such as precision, recall, and f1-score. As a result of the analysis, the model that demonstrated the best results for vehicle classification tasks was selected. A quantitative evaluation of the detector used subsequently was also conducted, confirming the effectiveness of the selected model. The third section describes the practical part of the study, which involved augmenting and expanding the dataset. After further training the model on the improved data, the classifier was integrated with the YOLOv11 detector. A web interface was implemented that provides convenient interaction with the system, allowing users to upload videos, view detection and classification results in real time, and analyze statistical data. Testing the system on real-world video surveillance data confirmed the effectiveness of the approach, although it revealed the need for further optimization for complex camera angles and lighting conditions.

A problem of control assignment of works with a large dynamically replenished (changed) volume of them is considered. Despite existing tools for automating large-scale software development (AGILE products, Waterfall, and others), the challenge of promptly and correctly intervening in their production process under changing external and internal conditions remains acute due to the resulting negative consequences (project delays, budget overruns, and unfinished projects due to unsatisfactory quality). The most pressing issue in large projects is the optimal assignment of work in accordance with the established current competencies of employees. However, several studies (such as surveys from Scrum Inc.) have noted that the relationship between the implementation of the most popular Agile products and increased software development efficiency is inconsistent. For this reason, developing improvements to existing tools is relevant for managing the production of large software projects, where interrelated resource constraints time, cost, and computational are particularly acute. The objective of this study is to present a model and its implementing algorithm for optimizing the work assignment process with a focus on big data (from the perspective of the developer), resulting in exponential gains in time and resources when comparing options in the form of alternatives (job, worker). A distinctive feature of the model is an optimal algorithm for ranking a large number of options, executed in real time, and resulting in a virtually unimprovable correct solution to the problem of sorting a dynamically replenished set of alternatives, which corresponds to the most important principle of “here and now” development, consisting of an immediate response to changing conditions and customer requirements. The gain is achieved through a non-mechanical combination of three methods: the classical paired comparison method, its modification, correct in the sense of K. Arrow's axiom on the independence of choice (preferences) from previously achieved rankings, and the selected algorithm for sorting the numerical sequence. Examples of numerical simulation are provided, confirming the declared characteristics of the algorithm for selecting the optimal alternative (job, worker) in a big data environment.

This paper addresses the problem of constructing time-optimal paths for a moving vehicle, specifically an autonomous underwater vehicle operating in an aquatic environment. The main difficulties stem from spatially non-uniform currents and the presence of both static and moving obstacles. We propose an approach based on the optical-geometric analogy and the Fermat–Huygens principle. The problem is formulated as a generalized eikonal equation that describes the propagation of a wavefront in a medium with a prescribed velocity vector field. This formulation reduces the original variational problem to a partial differential equation, eliminating the need for an exhaustive search over candidate trajectories. The arrival time field is obtained as the solution to a boundary value problem, and the optimal path is recovered by moving against the gradient of this field. We develop two algorithms for numerical implementation. The first, based on the fast marching method, calculates the minimal time field from the start point to all nodes of a computational grid. The second algorithm reconstructs the optimal trajectory from this field and extracts the section corresponding to a specified time horizon. We perform a series of numerical experiments on four test scenarios of increasing complexity: a uniform flow, two opposing flows, a flow with a vortex and static obstacles, and a scenario that combines moving obstacles with a vortex. In all cases, the algorithm constructs time-optimal paths successfully while ensuring safe obstacle avoidance. The scenario with dynamic obstacles demonstrated the ability to adjust the route dynamically. We compared the proposed approach with the Salp Swarm Algorithm (SSA). The results show that our algorithm finds paths with shorter travel times, even when their geometrical length exceeds that of the routes produced by SSA. Moreover, the computation time is sufficiently low to allow real‐time path planning.

The article addresses the pressing task of developing specialized methods for searching and ranking rubber compound formulations with similar compositions in databases. The aim of the research is to develop and conduct a comparative analysis of similarity metrics adapted for quantitatively assessing the proximity of multi-component compositional data represented as normalized vectors of ingredient weight fractions. The core of the work includes a formal statement of the problem of identifying relevant formulations, which requires maximizing a comprehensive similarity function that considers both qualitative composition (presence of ingredients) and quantitative proportions. Four metrics are proposed and adapted as tools: weighted Jaccard and Dice coefficients, Hellinger similarity, and cosine similarity. A theoretical analysis of their properties is supplemented by empirical validation on a real industrial database containing 6,096 unique formulations. The scientific novelty of the study lies in the systematic application and adaptation of metric analysis apparatus to the task of searching for analogues for compositional materials science data, as well as in revealing a fundamental clustering of the considered similarity measures. Unlike existing approaches focused on binary representation of composition or property prediction, the presented methodology purposefully solves the problem of precise search by composition and proportions. The obtained results revealed a near-functional equivalence of the weighted Jaccard and Dice coefficients (correlation coefficient r=0.991), forming one cluster of measures sensitive to the full set of components. Hellinger similarity and cosine similarity demonstrated a strong correlation (r=0.883), forming a second cluster of measures focused on assessing structural similarity of proportions, with the Hellinger metric showing increased sensitivity to variations in the fractions of minor ingredients. Based on this, practical recommendations are formulated for the combined use of one metric from each cluster to create effective search systems. The developed metric framework establishes a formal basis for intelligent analogue search, automation of component selection, and reduction of development time for new formulations in the industry.

Ensuring industrial safety at thermal power facilities requires effective and reliable early detection of potentially fire-hazardous and emergency situations. This task is complicated by specific operating conditions characterized by a wide range of operating modes, high levels of electromagnetic interference and vibration, and often a limited amount of labeled data suitable for training and verifying monitoring and forecasting systems. The goal of this study is to develop and experimentally validate a comprehensive approach to improving the accuracy and reliability of computer vision systems for monitoring thermal power facility safety based on an active learning methodology. The primary methods utilized include convolutional neural network models for object detection based on the YOLOv8 architecture, a heuristic mechanism for selecting informative data based on an uncertainty metric, and tools for explainable artificial intelligence and fault-tolerant decision-making with expert participation. The scientific novelty of the work lies in the adaptation of Active Learning methods to single-pass detectors for industrial safety tasks, as well as in the integration of a cycle of targeted additional labeling of complex examples with mechanisms for interpretation and hybrid verification of results (which allows for minimizing labeling costs (due to a selective request for "complex" examples); prompt adaptation to new types of anomalies (e.g., non-standard leaks, local overheating)); whereas most publications on the safety of thermal power facilities use classical machine learning methods (SVM, random forests) or pre-trained CNNs without adaptation to the specifics of the industry. In the completed work, experimental studies on a specialized dataset of images of thermal power facilities showed that the use of active learning provides an increase in the mAP@0.5:0.95 indicator by 28.32% and an increase in the detection recall by 7.82% compared to standard learning. The obtained results confirm the potential of the proposed approach for practical application in monitoring and early warning systems capable of ensuring the timely detection of deviations from normal operating parameters of equipment and preventing the development of emergency scenarios.

The article discusses the main components of digital twins and their interconnections, the specific features of distributed generation units, and the challenges that arise during their use. Digital twin technology is currently widely used in industrial production. Digital twins act as virtual analogues of physical objects, groups of objects, or processes. They are complex software solutions based on large volumes of data and the synthesis of artificial intelligence, machine learning, and specialized software, enabling the creation of dynamic digital models capable of live interaction and adaptation. The article presents a methodology for constructing digital twins of distributed generation units, the use of which helps optimize the design, operation, and monitoring of electric power systems. The methodology is presented using a turbogenerator unit as an example and includes the following stages: collecting data on the physical object, modeling the turbogenerator unit, and comparing experimental data and the response of a fuzzy model. The scientific novelty of the study lies in the development of a hybrid approach combining neural networks, a fuzzy inference system, and a genetic algorithm. The study identified the minimum root-mean-square deviation (RMS) of the model's response to experimental data, determined the optimal number of rules in a fuzzy system (144 rules), and established the relationship between model accuracy and computational costs. The practical significance of the results lies in the potential application of the developed method to optimize the operation of distributed generation units, improve power supply reliability, and enhance the management efficiency of electric power systems. The proposed approach enables the creation of accurate digital models capable of adapting to changing operating conditions. The research findings can be used in the development of new energy systems, the modernization of existing facilities, and the implementation of Smart Grid technologies. The proposed approach opens new prospects for the development of energy-efficient and reliable power supply systems.

This article presents an analysis of the energy efficiency of semiconductor direct current converters (DC/DC converters) used in photovoltaic installations. The study of power losses in the switches of a boost converter with pulse-width modulation (PWM) was carried out using a simulation computer model developed in the MATLAB environment with blocks from Simscape. The results of the simulation modeling of the graphs showing the dependence of output current and voltage on step changes in solar irradiance are presented. It has been established that with an increase in solar irradiance, the current and voltage increase, indicating a direct dependence of the generated current and voltage on the level of illumination. The developed model allows the determination of static and dynamic power losses in power semiconductor switches depending on the switching frequency. The analysis of the study showed that the DC-to-DC semiconductor converter with PWM control experiences significant losses during the switching on and off of semiconductor switches, which leads to a reduction in the efficiency of the converter. To reduce dynamic losses and increase the efficiency of the semiconductor converter, it is necessary to investigate a new control method based on the application of frequency- pulse modulation (FPM).

This article focuses on optimizing an enterprise's personnel security program using discrete (0-1) linear programming methods under resource constraints. This approach is relevant because personnel are both a key asset and a potential source of internal threats, ranging from unintentional errors to deliberate violations leading to financial losses. The objective of this study is to develop a formalized model for selecting a set of personnel and information security measures that, given budget and labor limits, ensures the required reduction in integral risk and maximizes the expected economic impact. Binary indicators for program inclusion are used as decision variables, and the optimality criterion is defined as maximizing net annual savings (the difference between prevented damage and costs). The model incorporates constraints on funding, available man-hours, and an optional minimum cumulative impact constraint. Initial data was generated for six alternative measures (DLP/StaffCop monitoring, UEBA analytics, MFA enhancements for privileged users, information hygiene e-learning, KPIs/early warning, and an expanded benefits package) indicating cost, labor intensity, and expected loss reduction based on expert assessment and incident statistics. Practical testing was performed in the Python environment using the MILP (branch-and-bound) approach and demonstrates the optimal set of measures. The resulting solution ensures compliance with resource limits and the achievement of the target effect, while eliminating measures with the worst cost-to-benefit ratio. The scientific and practical significance of this work lies in the translation of high-quality management reasoning about personnel security into a reproducible optimization formulation suitable for recalculation with changes in prices, labor resources, and regulatory requirements.

This article systematizes modern methods of intelligent career guidance and proposes a unified computational model of recommendations that integrates psychometric data, digital footprints, academic indicators, and textual descriptions of professions. The relevance of the study is driven by the transition from isolated diagnostic techniques to comprehensive, data-driven career support systems capable of accounting for heterogeneous user information and the dynamics of professional trajectories. The aim of the work is to enhance the quality and interpretability of career guidance recommendations by analyzing existing approaches and developing a mathematically rigorous model that integrates three key tasks: reconstructing professional interests, multimodal matching between user preferences and occupational profiles, and ranking possible career trajectories. The article presents an overview of studies published between 2020 and 2025, demonstrating improvements in the accuracy and robustness of career guidance methods achieved through ensemble algorithms, multimodal deep learning architectures, and LLM-based conversational systems. The analysis shows that combining psychometric profiles, digital activity, and textual descriptions of professions substantially increases the quality of recommendations in both interest reconstruction and career selection tasks. The scientific novelty of the work lies in proposing a unified theoretical and computational framework for intelligent career guidance, which provides a common mathematical structure for diverse existing approaches and enables viewing them as special cases of a general model. The proposed framework includes a formal representation of user and profession data, a RIASEC profile reconstruction function, a parameterized multimodal matching mechanism, and a joint optimization objective that simultaneously trains all components. This formalization establishes a foundation for developing interpretable, reproducible, and scalable career guidance systems.

The developed algorithms for solving nonlinear optimal control problems and approximating non-convex reachable sets formed the basis for specialized software constructed for studying nonlinear controlled dynamical systems. Such systems arise in various scientific, technical, and industrial fields and are characterized by a high degree of complexity (nonlinear dynamical systems and non-convex objective functionals). Therefore, their effective solution requires the use of algorithms that take into account the specific characteristics of these problems. The paper proposes appropriate computational technologies based on both classical approaches based on sequential discretization of continuous problems and the application of Pontryagin's maximum principle, as well as on the specific properties of the dynamical systems, such as the linear connectivity of the reachable set and the hidden convexity of the admissible velocities set of controlled dynamical systems. Pre-optimization analysis methods (estimating the degree of the objective functionals convexity and constructing the boundary of the reachable set) are also implemented as programs that allow for the initial assessment of the computational complexity of applied non-convex optimization problems and the selection of effective numerical methods for their solution. Descriptions of the mathematical, software, and technological formulations of studied non-convex optimal control problems are provided. A framework for synthesizing multi-method non-local algorithms for optimizing controlled dynamical systems is presented.
The stages of constructing the computational scheme and the specifics of selecting algorithmic parameter values are described. A developed test collection of nonlinear optimal control problems is used to test software implementations of non-convex optimization algorithms to study their limiting properties and find effective modifications. The collection includes both published problems with known solutions and problems generated using the proposed test generation methodology. The developed algorithms and corresponding software were used to solve practical problems in various scientific and technical fields: flight dynamics and space navigation, quantum physics and computational chemistry, synthesis of composite structures, economics, medicine, technical ecology, and other areas.

Currently, virtual assistants, or intelligent assistants, are actively used in various subject areas, but their widespread use in solving technical diagnostic problems is limited. Such limitations are mostly related to the plausible nature of the recommendations generated. One of the ways to increase the reliability of recommendations is to use classical artificial intelligence methods, in particular, those that implement reliable inference based on logical rules. The paper describes the process of building problem-specific assistants using declarative knowledge bases in the form of specialized decision tables to solve the problems of diagnosing technical systems. An original approach based on visual modeling and model transformations is used as a methodological basis. A generalized algorithm for diagnosing technical systems is presented, which includes steps to specify the object under study, the external manifestations of the malfunction, the search for possible causes (candidate systems for the malfunction), as well as the formation of a list of troubleshooting activities. The algorithm assumes the use of declarative knowledge bases containing logical rules of three types: "manifestation-system", "operation-operation", "case". The general architecture of assistants is described based on the concept of a template with sockets (slots) for connecting knowledge bases with rules of a certain type. The requirements for the structure and content of knowledge bases are presented, as well as ways to manage them. The technology stack is defined: Aimylogic JUST AI – for designing a dialog; PHP – for implementing software interfaces for accessing knowledge bases; CSV format – for describing logical rules in the form of decision tables. The application of the described provisions was carried out when creating a prototype of the intelligent assistant "Aviatech.Assistant" to support the technical specialists of airfield ground services. Knowledge bases have been created for individual Sukhoi Superjet (RRJ-95) and Cessna (182T) systems. A special feature of the resulting solution is its compliance with the principles of the declarative programming paradigm, which allows one to reconfigure or adapt it depending on the type of aircraft or documentation version without complete regeneration or recompilation.

This paper presents a solution for creating an autonomous microclimate monitoring system for technical facilities, capable of promptly detecting anomalies under conditions of non-stationary data and limited sensor infrastructure. The research aim is to develop monitoring tools that ensure continuous control, anomaly detection, and their interpretation. The core content of the work includes the development of a functional system model, the implementation of telemetry collection and heterogeneous data storage modules, and the deployment of an original hybrid algorithm for time series analysis. The key element is the proposed method, which combines spectral analysis to extract periodic signal components with adaptive statistical thresholds dynamically calculated based on sliding historical windows. Unlike existing counterparts relying on resource-intensive deep learning, the proposed approach utilizes physically grounded features to adapt to seasonal data variability without a preliminary training stage on labeled datasets. To unify the anomaly detection approach and alert routing, specific alert rules were configured, integration with a data visualization system was performed, and a chatbot was created to handle notifications and queries. To enhance fault tolerance, a component liveness control mechanism was implemented. The scientific novelty lies in proposing a hybrid adaptive anomaly detection method based on spectral-statistical time series analysis. The proposed approach ensures result interpretability by revealing the physical nature of failures through signal decomposition into harmonic components. Verification results on a real dataset of 5.3 million records confirmed the approach's effectiveness: detection recall reached 0.991 with a precision of 0.877. Load testing demonstrated the architecture's stability under low CPU utilization (<6%) and indicated reserves for scalability. The project successfully identified incidents related to air conditioning system malfunctions. The developed system provides a full cycle from data collection to preventive alerting, offering understandable diagnostic tools for preventing emergency situations while supporting two levels of interaction: a simplified interface for end-users and advanced analytical functions for specialists.