Moscow Aviation Institute

This paper explores the transformation of competitiveness in Russia’s electrical engineering sector under sanctions, digitalization, and the integration of artificial intelligence. It argues that competitiveness is no longer a linear function of product quality but a function of managing uncertainty. New metrics, strategies, and case studies are presented, highlighting the paradoxical role of risk as a resource.

<p><strong>Context and relevance</strong>. The tasks of synthesizing optimal control for nonlinear dynamic systems with constraints and uncertainties remain computationally challenging, especially in aerospace applications. Reinforcement learning is considered a practical tool for building feedback and/or accelerating planning when classical methods are difficult to apply. <strong>Objective</strong>. To systematize classes of algorithms for optimal control tasks and identify criteria for selecting an approach for a specific problem. <strong>Hypothesis</strong>. Practical applicability is ensured by correct formulation and consideration of requirements for control continuity, data, safety, and robustness; combined solutions are most effective. <strong>Methods and materials</strong>. A review and comparative analysis of families of different reinforcement learning algorithms was performed. <strong>Results</strong>. Actor-critic remains the basis for continuous control, while alternatives increase selective efficiency but are sensitive to model and data coverage errors. <strong>Conclusions</strong>. The most promising are hybrid architectures that combine reinforcement learning with basic controllers and ensure controlled compliance with constraints. The choice of method should be determined not only by quality, but also by safety, robustness, and computational cost.</p>

Today, the public sector is facing dynamic technological changes, with AI being among the most innovative areas. The popularization of relevant IT solutions and their widespread use in business have led to growing interest in this technology within public administration and local self-government. This trend is evident in many countries, and the introduction of AI-based solutions in local self-government bodies may have a stronger effect on achieving development principles. The study was conducted using a mixed design: a systematic literature review (2018-2025) and an expert assessment with ranking (n=43, inclusion ≥3 peer-reviewed publications). Experts were asked to rank the proposed directions. The ranks were converted into normalized weights using the rank sum method, and the consistency of assessments was verified with Kendall’s coefficient of concordance (W=0.64-0.67; p<0.01). The most prioritized areas of AI application are the management of large data sets and resource allocation; the main barriers are high costs, a shortage of competences, and insufficient IT tools; the key threats are risks to personal data protection, violations of intellectual property rights, and cybersecurity issues. Practical recommendations are provided on the stages of implementation and risk minimization.

Methods for constructing complex flight trajectories of aircraft are analyzed, and the specific features of their formation are examined. A universal algorithm is proposed for generating an arbitrary spatial trajectory for a maneuvering aircraft in simulation tasks. The algorithm involves forming a sequence of motion segments that implement elementary maneuvers. Possible variants of the algorithm's construction, as well as limitations and extensions in its implementation, ar discussed.

A mathematical model for credit debt repayment by a representative rational household has been developed. The model is formalized as an optimal control problem with a noncompact geometric control constraint and a nonfixed time horizon. An existence theorem for solutions in the class of functions admitting impulse control has been proved. Necessary optimality conditions in the form of Pontryagin’s maximum principle have been investigated. Optimality conditions for the early debt repayment regime have been derived.

The aim of this study is to develop an acoustic emission (AE) monitoring methodology that is invariant to the mechanical impact influence. To achieve this objective, regression and statistical analysis algorithms were employed to identify nonlinear relationships between the degree of damage in composite specimens (j) and the values of streaming AE parameters. As the most informative AE parameters correlating with the actual condition of the monitored specimens, the quantile values at the p = 0.9 level of the distribution functions of the front energy ([Eφ]p = 0.9) and specific energy ([EN]p = 0.9) of AE pulses were selected. A key stage in the implementation of the proposed methodology involves partitioning the criterion plane ([EN]p = 0.9 – [Eφ]p = 0.9) into three characteristic segments (I, II, III). Based on the values of the streaming parameters and their weighted distribution across segments I, II, and III (WI, WII, WIII), a regression model was synthesized to assess the degree of damage in composite specimens using AE monitoring results. The maximum reduced error of the developed model was γ = 9.8%, while the mean relative error did not exceed 1.1%, regardless of the initial condition of the monitored specimens.

Aim is to improve the physical development and physical fitness of primary school children who are overweight through physical education. Materials and methods The study was conducted from January to May 2024 at the comprehensive school in Moscow (Russia). The pedagogical experiment involved schoolchildren who study in grades 1-4 who were diagnosed with obesity of varying degrees (72 children). Children in the experimental group did physical education with a separate instructor according to the experimental program, and schoolchildren from the control group studied in one with the rest of their class. Before and after the pedagogical experiment, all schoolchildren completed control tests that assessed their level of physical development and physical fitness. Results During the study period, the schoolchildren from the control group showed an increase in the indicators of less than 2% in the EVC, Height, Weight, BMI, Jump and Run tests (p> 0.05). In the RI test, the indicators improved from 10.3±1.6 to 9.4±0.65 (t=0.52; p>0.05), the increase was 9.6%. In the Torso test, the indicators improved by 11.5% (p> 0.05), and in the Jump Rope test, the result was higher by 14.6% (p> 0.05). In the experimental group, a reliable increase in the results was recorded in 7 of 9 tests. The smallest increase in results was noted in the Height test - 1.1% (p>0.05) and Run - 0.6% p>0.05. The EVC increased by 14.2% (p<0.05), the Weight decreased by 7.8% (p<0.05), the BMI decreased by 9% (p<0.05), in the Jump test the result improved by 9% (p<0.05). In the RI test the result improved from 10.9±1 to 7.1±0.75 (t=3; p<0.05), the increase was 53.5%, in the Torso and Jump Rope tests the indicators improved by 31.4% (p<0.05) and 38.3% (p<0.05), respectively. Conclusion The results of the study in the control group indicate the low efficiency of the standard physical education program in a comprehensive school for children aged 7-11 who have excessive weight; the program is more suitable for children who do not have health limitations. The results of the tests in the experimental group indicate the efficiency of using the experimental program for children aged 7-11 who have excessive weight. These schoolchildren should do physical education according to a special program and perform physical exercises that are feasible for them. BJMS, Vol. 25 No. 01 January’26 Page : 253-262

This article focuses on the development and investigation of the matrix-based modifications of the Ant Colony Optimization (ACO) method for solving parametric optimization problems using CUDA-based Graphics Processing Units (GPUs). Unlike most existing studies focusing on the Traveling Salesman Problem (TSP), the proposed method is specialized for finding optimal values of independent parameters, which leads to a different graph structure and requires the development of novel parallelization schemes. Original parallelization strategies are proposed, including a three-stage decomposition of the algorithm and introduction of “parameter-level” parallelism in the problem dimensions. Various strategies for parallelization and memory management were developed and tested: from the basic MatrixACO-Basic modification to the optimized versions of MatrixACO-Optimization (featuring warp reduction and optimized hash-table handling) and MatrixACO-Transported (utilizing transposed data storage). Experiments were conducted on NVIDIA Tesla V100 16GB SXM2 GPU, the Jetson ORIN platform, and various personal computers using multi-dimensional test functions. Results demonstrate that for a problem with 128 parameters the optimized version of MatrixACO-Optimization provides an execution time of 0.13 ms per graph layer, which corresponds to a speed-up of more than 30 times compared to the optimized CPU implementation based on OpenMP, and in the case of a very large-scale problem with 65536 parameters it speeds up by almost 45 times. For a problem with 512 parameters, the GPU realization shows 76% higher energy efficiency compared to the CPU version, as well as an almost eightfold (7.9 times) reduction in Energy-Delay Product. The study confirms the necessity of developing specialized GPU implementations for parametric tasks and reveals the excellents calability of the proposed algorithms. Статья посвящена разработке и исследованию матричных модификаций метода муравьиных колоний (ACO) для решения параметрических задач оптимизации с использованием графических процессоров (GPU) на архитектуре CUDA. В отличие от большинства существующих исследований, фокусирующихся на задаче коммивояжера (TSP), предложенный метод специализирован для поиска оптимальных значений независимых параметров, что приводит к иной структуре графа и обуславливает необходимость разработки новых схем распараллеливания. Представлены оригинальные схемы параллелизма, включающие разделение алгоритма на три этапа и введение “параметрического” параллелизма по измерениям задачи. Разработаны и протестированы различные стратегии распараллеливания и управления памятью: от базовой модификации MatrixACO-Basic до оптимизированных версий MatrixACO-Optimization (с warp-редукцией и оптимизацией хэш-таблицы) и MatrixACO-Transported (с транспонированным хранением данных). Эксперименты проводились на графических ускорителях NVIDIA Tesla V100 16GB SXM2, платформе Jetson ORIN и различных персональных компьютерах с использованием многомерных тестовых функций. Результаты показывают, что для задачи с 128 параметрами оптимизированная версия MatrixACO-Optimization обеспечивает время обработки одного слоя графа, равное 0.13 мс, что соответствует ускорению более чем в 30 раз по сравнению с оптимизированной CPU-реализацией на основе OpenMP, а в случае задачи с 65536 параметрами способствует ускорению почти в 45 раз. Для задачи с 512 параметрами GPU-реализация демонстрирует на 76% более высокую энергоэффективность по сравнению с CPU-версией, а также почти восьмикратное (в 7.9 раза) снижение Energy-Delay Product. Исследование подтверждает необходимость создания специализированных GPU-реализаций для параметрических задач и выявляет отличную масштабируемость предложенных алгоритмов.