Electronics and Control Systems

Multi-agent Deep Reinforcement Learning in the Collision Avoidance Problem

2026-03-13T13:44:38+02:00

Obstacle avoidance is crucial for the succesfull completion of unmanned aerial vehicles missions. This article is devoted to the research of the multi-agent deep reinforcement learning in the collision avoidance problem. It is considered unmanned aerial vehicle swarms encounter diverse obstacles categorized into: static large-scale and small-scale obstacles, dynamic large-scale and small-scale obstacles, complex terrain, thin/low-visibility obstacles, partially-occluded/transparent obstacles. To address the abvove problem, a multi-agent deep reinforcement learning based trajectory control algorithm is proposed for managing the trajectory of each unmanned aerial vehicle independently. It was researched different approaches and multiple 3D simulation environments with help of reinforcement learning for the swarm of unmanned aerial vehicles.

An Integrated Techno-economic Framework for SMS Delivery Optimization in 4G/5G Networks

2026-03-07T22:54:56+02:00

Short Message Service (SMS) remains a critical signaling-layer communication mechanism in 4G and 5G mobile networks, particularly within Application-to-Person (A2P) ecosystems supporting authentication, financial notifications, public services, and IoT fallback signaling. The transition to LTE packet-switched delivery and further to 5G Service-Based Architecture (SBA) introduces dynamic routing, SLA differentiation, multi-channel fallback, and cost variability. Traditional Quality of Service (QoS) indicators are insufficient to assess operational efficiency in such environments. Furthermore, Artificially Inflated Traffic (AIT) distorts economic performance without necessarily degrading technical KPIs. This paper proposes an integrated techno-economic framework for SMS delivery optimization in next-generation networks. The framework introduces the Price Delivery Gap (PDG) as an economic deviation metric and develops the Integrated Gap-Delivery-Performance (IGDP) model combining QoS, Quality of Experience (QoE), and PDG into a unified optimization function. An intelligent architecture incorporating message categorization, AIT detection, adaptive routing, and closed-loop monitoring is presented. Scenario-based evaluation under mixed A2P traffic conditions demonstrates reduced economic deviation and improved integrated efficiency compared to static and partially adaptive approaches.

Determining the Effectiveness Criteria for Intelligent Detectors in Integrated Video Surveillance Systems

2026-03-07T22:26:24+02:00

The paper considers a comprehensive methodology for evaluating the effectiveness of video surveillance systems based on classical motion detection and neural network object recognition algorithms. A system of technical, financial, and operational performance indicators is outlined, including event miss probability, recognition precision, false alarm rate, response time, infrastructure costs, staffing expenses, and operator workload index. The integral detection quality metric based on the F1-score, defined as the harmonic mean of Precision and Recall, is characterized. A weighted aggregation model is proposed that accounts for the criticality level of the protected facility and enables adaptive balancing between technical and economic factors. The mechanism of incident cost impact on financial efficiency through expected losses from missed events is analyzed. Comparative calculations are presented for facilities with different camera counts (10 and 200) and different criticality levels. It is established that under high-criticality conditions neural network algorithms demonstrate significantly higher overall efficiency compared to traditional motion detectors due to lower miss probability and reduced operator workload.

Topology Management of a Swarm of Unmanned Aerial Vehicles

2026-03-07T19:21:32+02:00

The article is devoted to the problem of controlling the topology of a swarm of unmanned aerial vehicles (UAVs). The primary objective of swarm operation is maintaining a dynamic topology, i.e., stable information exchange and structural consistency between swarm elements in a constantly changing environment. It is shown that existing approaches rely on a global navigation satellite system (GPS) for UAV positioning, such as the Global Positioning System (GPS). This approach is unacceptable, as UAVs can suddenly experience loss of GPS signals while performing their missions, potentially resulting in a lack of location information. To support functionality, including the use of geographic routing protocols, maintaining connectivity in dynamic network conditions, and adapting to topological changes in the UAV swarm, this paper utilizes virtual coordinates. This paper develops a virtual coordinate system that forms the basis of the proposed method for UAV swarm topology management. This eliminates the need for global coordinates, a centralized controller, motion model coordination, and pre-calibration of the swarm formation. The system operates solely based on local distance measurements to neighbors, making it universal, scalable, and resilient to the loss of individual UAVs. Algorithms for merging and separating swarms based on the virtual coordinate system have been developed.

Investigation of Arrival Operational Dynamics and Safety in Point Merge System

2026-02-28T05:42:08+02:00

The paper addresses the critical issue of enhancing Air Traffic Management efficiency in terminal manoeuvring areas amidst recovering flight traffic intensity. The study focuses on the operational stability of arrival flows, specifically comparing the effectiveness of the innovative Point Merge System (PM) against traditional Radar Vectoring. The research methodology is based on the analysis of large-scale real-world ADS-B trajectory data acquired from the OpenSky Network for Dublin Airport (EIDW). An ETL (Extract, Transform, Load) approach was applied, and a Python-based software suite was developed to calculate Key Performance Indicators, including arrival headway stability, indicated airspeed variability, and trajectory efficiency. The computational experiment results demonstrated that the geometric structure of PM functions as a "passive controller," transforming the stochastic arrival process into a deterministic closed-loop system. It was established that PM implementation reduced velocity variability in the sequencing zone by an average of 12–15 knots, minimizing the "accordion effect" and flight crew workload. Analysis of the Empirical Cumulative Distribution Function (ECDF) for headways revealed a significant reduction in standard deviation, indicating the dissipation of operational entropy. Although the average flight distance within the PM system increased by 12.4%, this is offset by a 60% reduction in holding time. The study concludes that the PM System provides a superior level of predictability and safety margins during peak loads, effectively trading minor distance extensions for enhanced flow stability.

Docking of UAV for Air-to-Air Refueling Under the Influence of the Bow Wave

2026-02-22T20:18:01+02:00

The issues of automating the air-to-air refueling of unmanned aerial vehicle are considered here. The focus is on the contact phase of the "floating-up" drogue with the refueling probe of the refueling unmanned aerial vehicle. The paper discusses affairs of formation of a rendezvous trajectory using a laser beam from a tanker's gyrostabilized optoelectronic system. In this laser beam must keep the refueling unmanned aerial vehicle and the actively controlled refueling drogue. To eliminate the unpredictability of the drogue's "floating-up" direction during the contact phase, the "offset aiming" strategy and algorithms for countering drogue displacement caused by the bow wave effect are proposed. An optimal contact trajectory in terms of approach speed is proposed. The proposed algorithms and their modifications were investigated using mathematical modeling. Studies have shown that the proposed algorithms for compensation of the drogue's "floating-up" effect are quite workable.

Synthesis of a Regulator for UAV Motion Control System under Uncertainty

2026-02-22T19:47:45+02:00

This article represents the study of uncertainties inherent in motion control systems for unmanned aerial vehicles. Both external and internal disturbances acting on moving objects are considered. Expressions for the turbulent wind are given. The analysis of structured and non-structured uncertainties is presented. Results of studying different types of regulators are given, including PID regulators, LQR regulators, robust regulators, and regulators based on non-linear approaches and artificial intelligence methods. The procedure of H infinity synthesis is described. The block diagram of the conversion a continuous regulator in discrete one is represented. The comparative analysis of application LQR and H infinity regulators in loops of tracking by a given trajectory for conditions of the normal and disturbed atmosphere is given. The appropriate graphical dependencies are shown. The obtained results can be useful for aerial objects of a wide class.

Urban LoRaWAN Resilience under EW Interference: an Operational Model for Municipal Networks

2026-02-22T19:01:06+02:00

This paper reports field experience from a municipal LoRaWAN network that periodically faces electronic-warfare interference. We connect the observed service failures to a simple SINR-based reception model and translate that model into practical actions an operator can take – without expensive instrumentation and without heavy changes on end devices. The result is an operations-oriented playbook: which radio frequency symptoms to watch for in gateway / network-server data, how to recognize likely interference patterns, and what mitigations help keep core city services running under degraded radio conditions.

Prediction of Vestibular Schwannoma Growth Based on Radiomics Features of MRI Images Using Ensemble Machine Learning Methods

2026-02-21T18:46:11+02:00

This paper proposes a method for predicting vestibular schwannoma growth based on the analysis of a single MRI scan using radiomics features and ensemble machine learning methods. A total of 96 patients from the public Vestibular-Schwannoma-MC-RC2 dataset were studied. 744 texture features were extracted using wavelet decomposition. A Voting ensemble combining five classifiers was proposed: SVM, logistic regression, k-NN, Random Forest, and LDA. ROC AUC of 0.742 ± 0.072 was achieved using 5-fold cross-validation. The results confirm the effectiveness of the proposed approach for early prediction of tumor growth.

Methodology of Designing Systems for Spatial Stabilization

2026-02-21T18:08:09+02:00

This article presents a study of the design features of spatial stabilization systems. The described methodology covers basic methods and procedures necessary for the creation of a precision, resistant to disturbances, and at the same time controllable system for precision stabilization of the equipment assigned for operation on moving vehicles of a wide class. The block diagram explaining the interconnection between basic methods and procedures for designing spatial stabilization systems is represented. The algorithms for choosing the controllable system structure and the synthesis of the robust controller are described. The procedures for the choice of the system’s components and modelling of both system components and the system as a whole are presented. The features of the modern approach to data processing based on neural networks are described. The obtained results can be useful for the spatial stabilizing objects assigned for operation on moving objects of a wide class.

Definition of MicroGrid Features for the Construction of an Intelligent Control System

2026-02-21T17:37:03+02:00

The article addresses the problem of identifying the key features of a MicroGrid as a complex energy system, which is a necessary prerequisite for the development of an intelligent control system. The widespread introduction of distributed generation, renewable energy sources, and energy storage systems necessitates adaptive control methods capable of operating under conditions of uncertainty and variable electrical network modes. Existing approaches to MicroGrid control remain limited due to the lack of a unified and scientifically grounded set of structural, operational, dynamic, and informational features that should be taken into account in intelligent systems. The paper analyzes contemporary scientific approaches to the classification, modeling, and real-time control of microgrids, which makes it possible to identify insufficiently studied aspects related to the formation of feature sets and their influence on decision-making processes. The objective of the study is to systematize and substantiate a comprehensive set of microgrid features that can serve as a basis for the development of intelligent control algorithms. A conceptual approach to structuring microgrid features into the following groups is proposed: architectural, energy-technical, operational, features related to the stability of dynamic operating modes, and data-oriented features. Their significance for forecasting, optimization, state estimation, and autonomous operation of local energy systems is demonstrated. The obtained results form a methodological foundation that enhances the effectiveness of developing intelligent microgrid control systems and contributes to improving the flexibility and reliability of local energy networks.

A Model and Method of Transactional-behavioral Data Mining for B2B Content Personalization

2026-02-21T17:00:58+02:00

The growth of financial significance and structural complexity of the B2B e-commerce market segment, alongside the necessity to increase its efficiency, has determined the relevance of developing a model and method for the intelligent analysis of B2B customer transactional and behavioral data for content personalization. This study analyzes data mining methods based on Apriori, FP-Growth, and Eclat algorithms and data structures, identifying ways to improve them for B2B-specific data analysis. A conceptual model for analyzing B2B customer commercial activity has been developed, incorporating the product of item quantity and individual B2B transaction price. Furthermore, the UP-Growth (Utility Pattern Growth) method has been developed, utilizing a weighted node utility calculation within the tree structure instead of the standard frequency counters used in FP-Growth. The paper provides examples of constructed association rules and sequential patterns, accompanied by explanations of their economic significance. The impact of the derived association rules and sequential patterns on the formation of personalized product, information, and recommendation content within B2B e-commerce systems is examined.

A Method for Preparing Convolutional Neural Networks for Edge Deployment

2026-02-03T15:00:36+02:00

This study addresses the critical problem of accuracy loss during the compression of deep neural networks for mobile platforms. The research focuses on optimizing convolutional neural networks for operation under constrained hardware resources and the Memory Wall effect. An innovative Edge-deployment preparation method is proposed, which, unlike traditional sequential approaches, integrates structured pruning, post-training quantization, and a fine-tuning stage into a single iterative cycle. This approach provides a synergistic effect, minimizing accuracy degradation while achieving maximum parameter compression. Comparative analysis results confirm that the developed method meets strict latency and power consumption constraints, which are vital for mobile diagnostics in medical applications. Future research prospects involve adapting this method to other machine learning architectures.

Comparative Analysis of Satellite Images Stitching Methods Based on Local Feature Detection

2025-12-20T23:13:55+02:00

This paper investigates feature-based methods for satellite image stitching under a unified evaluation framework. Four algorithms – SIFT, SURF, ORB and BRISK - are examined with respect to keypoint detection, descriptor formation, correspondence generation and geometric alignment. A standardized MATLAB workflow is employed: grayscale detection and description, nearest-neighbour matching with a ratio test, robust outlier rejection via RANSAC with model escalation and mask-based blending with content cropping. Approximately fifty image sets spanning diverse landforms are processed; a Sahara Desert example illustrates the protocol. The study’s aim is to characterize the accuracy-efficiency trade-offs of vector (SIFT, SURF) and binary (ORB, BRISK) descriptors in realistic orbital conditions and to provide a transparent basis for method selection in remote-sensing workflows.

Definition and Intelligent Extraction of Texture Features of Vestibular Schwannoma Based on MRI Imaging

2025-12-20T22:17:48+02:00

The scientific work is devoted to the development of a method for intelligent extraction of textural features of vestibular schwannomas based on magnetic resonance imaging images for predicting tumor growth. The VS-MC-RC2 dataset was analyzed (421 timepoints, 189 patients, 1990–1999). The ML dataset consists of 211 samples (74 growing, 137 stable, imbalance 1.85:1). Gray Level Co-occurrence Matrix and Gray Level Size Zone Matrix matrices, shape features, wavelet transform, and the PyRadiomics v3.0.1 library were used to extract features from T1C images (priority) and T1 images (fallback) with the following parameters: bins = 32, δ = 1 voxel, 13 3D directions. Model v2 (107 original features) achieved an AUC of 0.618. Model v3 (851 features + 8 wavelet decompositions) achieved an AUC of 0.712 (+15.2%). Validation was performed using 10-fold cross-validation with an 80/20 train/test split. Among the top 15 features, 73% were wavelet features (LHH, LLH, HLH). The best feature, original_glszm_ZoneEntropy (F = 12.67, threshold = 4.51), correlates with the Antoni A/B tissue ratio and the proliferative activity of the tumor.

Reverberation Time Errors From Frequency Domain Filtering

2025-12-20T21:26:11+02:00

Information about the frequency dependence of reverberation time is essential for addressing several tasks, including mitigating the impact of reverberation on speech quality and intelligibility, as well as assessing intelligibility using the indirect modulation method. To obtain this information, the room impulse response must be filtered using a bank of octave or one-third-octave filters. This paper analyzes the influence of frequency bandwidth and the shape of the filter’s amplitude-frequency response on the bias of T₂₀, T₃₀, EDT, and T₁₀ estimates of the T₆₀ reverberation time. The analysis assumes that filtering is implemented in the spectral domain by zeroing the spectral components of the RIR outside the desired passband, and that the filter's amplitude-frequency response has the shape of a Tukey window. The results show that the use of filters with a rectangular amplitude-frequency response (Tukey window with parameter r = 0) is undesirable, as it leads to significant bias in the T₂₀, T₃₀, EDT, and T₁₀ estimates. This bias can reach 60–100% for reverberation times in the range of T₆₀ = 0.4–1.2 s. Using filters with a Tukey window shape and r = 1 reduces the bias to no more than 4% when filtering room impulse responses with octave filters at center frequencies f₀≥ 125 Hz. For one-third-octave filters with f₀≥ 25 Hz, a similar level of bias is observed for T₂₀ and T₃₀ estimates. For EDT and T₁₀ estimates, a bias of no more than 4% is achieved within the T₆₀ = 0.6–1.2 s range.

Influence of Airfoil Layouts on UAV Aerodynamic Characteristics at High Angles of Attack

2025-12-14T21:47:16+02:00

This article represents the study of the influence of airfoil layouts on aerodynamic coefficients at high angles of attack. The review of the previous research on the studied topic is given. The features of experimental equipment are described, including the set of sensors for the aerodynamic balance and the information and measurement system. The features of the described experimental equipment allow to realize the automation of the experimental test. The main features of the experiment technique are given. The main functions of the information and measurement system are listed. The results of the experimental test are represented as graphical dependencies of aerodynamic coefficients on angles of attack. The detailed analysis of the obtained results has been done. The results of the study can be useful for designing unmanned aerial vehicle motion control systems and simulating unmanned aerial vehicle motion, taking into consideration aerodynamic disturbances.

Unmanned Controlled Complex for High-quality Communication in Contemporary Extreme Medicine

2025-12-14T20:39:22+02:00

The general analysis was done of requirements for communication during of medical doctors work in extreme conditions, including locations of military activity. The tasks were determined – it was the need to create reliable stable mobile communication for their work; in the locations where there is no coverage, such communication can be organized using drones, and formation of complex of drones and ground base stations. So, the purpose was stated - to find solutions how to develop technologies to ensure high-quality communication with the use of unmanned aerial vehicles in extreme conditions in Ukraine. Numerous prototypes, including patents, were examined and analyzed. Several different schemes for organizing of mobile communications in cases where conventional wireless networks are vulnerable were observed (due to military operations, natural disasters, extreme situations). We have described some possibilities for the restoring of communications using a number of technological solutions based on the use of drones in extreme conditions. The possibility of software or hardware tools for replacing fixed base stations in wireless network with unmanned aerial vehicles were observed. It was proposed to use such tools in two ways: 1) either temporarily to replace fixed base stations with unmanned aerial vehicles; or 2) on permanent basis, f.e., in geographical locations where the terrain makes it difficult to place fixed base stations (impossible due to dangers of military operations, extreme situations, etc.) Such solutions can help to minimize also great damages of the war to our medicine, agriculture, nature in general.

Features of Mathematical Modelling of Gimballed Inertial Navigation System for Marine Moving Vehicle

2025-12-14T10:22:57+02:00

This article represents the features of creating the mathematical model and carrying out modelling of the gimballed inertial navigation system assigned for operation on marine moving vehicles. To increase the accuracy of the system, some modes of operation are introduced. Features of correction for every mode are described. The characteristic of the integral correction is given. The control moments for levelling and gyrocompassing modes are represented. The expressions for projections of the gyro-stabilized platform angular rates are created. The simulation results of stabilization and navigation processes are represented. The advantages of the integral correction are shown. The obtained results can be useful for the high-precision navigation systems and gyroscopic stabilization systems with payload. The proposed approaches can be applied for moving objects of the wide class.

Feature Extraction for Multispectral Analysis of Cereal Crops Using Optimized Computer Vision Pipelines 44

2025-12-14T09:54:37+02:00

The article presents the results of a study aimed at improving the stability, reproducibility, and structural consistency of computer vision pipelines for multispectral UAV imagery of winter wheat canopies. A new adaptive preprocessing model is introduced, incorporating illumination normalization based on a modified Retinex/MSRCR algorithm, entropy-regulated spatial-spectral filtering for noise suppression, and instability-driven spectral feature fusion to obtain stable multispectral descriptors. The model is formulated as a multi-objective preprocessing framework, jointly optimizing illumination invariance, noise robustness, structural preservation, and information richness. Experiments conducted on the open-access UAV dataset of nine winter-wheat fields (Switzerland) demonstrated a reduction of the coefficient of variation to 0.12 and RMSE to 0.089, together with improvements in structural similarity (SSIM = 0.923) and spectral entropy (H = 5.9), significantly outperforming classical normalization methods. The results confirm the effectiveness of the proposed approach in mitigating illumination heterogeneity and sensor-induced distortions, ensuring stable and phenologically consistent feature extraction. The developed framework can be integrated into computer-integrated and robotic precision-farming systems to enhance the reliability of automated monitoring and decision-support processes in winter-wheat production.

Computer Vision for UAV-based Reconnaissance under Conditions of Modern Warfare

2025-12-13T23:19:19+02:00

This paper considers the application of computer vision and deep learning methods for automated aerial reconnaissance using unmanned aerial vehicles under the conditions of modern warfare. The main classes of reconnaissance objects are analyzed, including military vehicles, fortifications, artillery positions, and groups of personnel. An approach to building an object detection system based on deep neural networks is proposed, in particular using YOLO-type detectors and U-Net segmentation models. The process of data preparation and augmentation with consideration of combat factors (smoke, explosions, low illumination, image shift, and noise) is described. An experimental evaluation of object detection quality under different scenarios is performed. It is shown that the use of specially adapted augmentation significantly increases the robustness of the models to interference. The limitations of the proposed approach and directions for further research are discussed.

Intellectual Diagnostics of Thyroid Tumors

2025-12-13T22:32:22+02:00

The article is devoted to the intelligent diagnosis of thyroid tumors, the diagnosis of papillary thyroid cancer based on general information, ultrasound images, and pathohistological images. It examines modern approaches to the intelligent diagnosis of thyroid tumors using machine learning and artificial intelligence methods. The types of medical intelligent systems, their architecture, accuracy, and the set of tasks they perform for the classification of thyroid cancer are considered. The problems of papillary thyroid cancer are considered, the specifics of the disease and the signs by which it is diagnosed are described. The main tasks of an intelligent system capable of automatically analyzing patient medical data and supporting clinical decision-making by an endocrinologist, segmenting and classifying thyroid tumors are outlined. The equipment used to form the training sample is described, and the process of data collection for building an intelligent medical system is described. The task to be solved is set. The metrics by which the accuracy of the intelligent medical system will be evaluated are characterized. The architecture of the intelligent medical system is presented, its main blocks are characterized, and the functionality of each block is described. A UML diagram is presented, according to which the intelligent medical system will operate. The data that will be used to form the training sample is presented, indicating its type, dimension, how the data is collected, and how this data will be used to train the intelligent medical system. The results of the study are aimed at improving the effectiveness of early detection of thyroid pathologies and reducing the number of false diagnoses, creating a convenient tool that will reduce the time it takes for a doctor to diagnose the disease and increase the accuracy of diagnosing papillary thyroid cancer.

Analysis of Methods for Monitoring the Condition of Building Facades Based on Visual Data

2025-12-13T17:03:11+02:00

The article explores the use of information technology to monitor the condition of building facades based on visual data obtained from unmanned aerial vehicles. The study highlights the growing role of unmanned aerial vehicles in structural inspections, noting their key advantages, including increased safety, efficiency, and accuracy, compared to traditional methods. The study is structured into three main sections. The first section provides an overview of existing approaches to facade monitoring, comparing traditional inspection methods with UAV-based methods. The second section discusses the technological aspects of data collection, processing, and analysis, focusing on artificial intelligence, computer vision, and photogrammetry. The final section presents the practical application of these technologies, an overview of relevant software tools, examples, and economic benefits. The results show that unmanned aerial vehicles, combined with advanced image processing technologies, significantly increase the efficiency and reliability of building facade assessments.

Technology of Objects Recognition by Drones with Humanitarian Medical Cargo

2025-12-11T19:56:28+02:00

The problems of objects recognition by drones (UAVs) were observed, the general analysis of these problems was done. The central purposes of the work were outlined, as well as the set of tasks for these problems solutions. In process of the work the observation of prototypes and analogs – the versions of previously constructed UAV modules with images recognition abilities were done for providing medical assistance in extreme conditions. The aim of present work was to develop mechanisms and possibilities of image recognition by UAVs for medical purposes, and to clarify the possibility of humanitarian cargo transportation by UAVs to programmatically defined object or person. To do this the experience of prototypes creation, and their analysis were taken into account. The algorithms of necessary operations and suitable software were developed to automate processes. Developed software was designed basing on a face recognition technology convolutional neural network. All necessary information concerning all stages of the work was described profoundly. Originally developed humanitarian UAV was constructed with various modules (for diagnosing a person's health state, providing patient with appropriate medical care, others); and module for images recognition in point of destination was one of them. The sample of created algorithm and few fragments of novel program supply were given in present article. Developed UAV can be used repeatedly for different tasks solutions: monitoring of area pollution (chemical or other); medical equipment and first aid means can be parachuted over the point of location of potentially injured person, etc. In both cases images recognition functions are desirably for the drone. Preliminary data concerning successful application of work results, some given practical recommendations were described.

Research on UAV Air Launch in the Jsbsim Software Environment

2025-12-11T19:28:38+02:00

The article considers the scientific and practical aspects of unmanned aerial vehicle air launch, analyzes modern approaches and technologies, and also determines the advantages of this launch method compared to traditional methods. The method of studying unmanned aerial vehicle air launch using simulations in the JSBsim software environment is considered. The database of unmanned aerial vehicle air launch dynamics is presented, confirming the effectiveness of using unmanned aerial vehicle air launch simulation in the JSBsim software environment. The influence of unmanned aerial vehicle weight on the launch process and subsequent unmanned aerial vehicle flight is considered in detail. The scientific novelty lies in the development of a method for high-precision modeling of unmanned aerial vehicle air launch in a circuit with an autopilot, the creation of a high-precision database of unmanned aerial vehicle air launch dynamics.

Air Traffic Controller Workload as a Factor in Multi-criteria Arrival Sequencing within the Point Merge System

2025-09-30T11:54:50+03:00

Gate-release strategy in the Point Merge System is crucial for reliable arrival sequencing and separation assurance in terminal areas. In this study, we examine three aggregation policies for the exit decision from the radius-to-fix arc conjunctive (AND), disjunctive (OR), and majority (MAJORITY) implemented with a non-compensatory safety barrier and S* speed-control variants. The objective is to assess each policy’s ability to regulate headways, maintain time-based separation, limit low-altitude level-offs, manage advisory demand, and mitigate environmental impact under varying weather conditions, and to identify their strengths and weaknesses. As an example, we conduct an experimental evaluation on the published geometry of Dublin (EIDW) RWY 28L, parameterising arrivals with realistic kinematics and stratifying by METAR; performance metrics include headways at arc/gate/final, spacing error relative to S*, a time-based loss-of-separation proxy, level-off time, and coarse terminal-area fuel / CO₂. Human factors are incorporated through a Human Workload Index combining expected speed-advisory count, level-off time, short-headway alarms, and weather difficulty markers. Alternatives are ranked using TOPSIS with AHP-like weights over Safety, Efficiency, Human, and Environment. The results show that policy choice is the primary driver of headway regularity, advisory load, and low-altitude behaviour; moreover, treating workload as a first-class criterion can overturn rankings obtained from an efficiency-only view. This evaluation helps practitioners select and gate-release policies to site-specific tolerances within an auditable framework.

About the Phase Noise of Frequency Synthesizers

2025-09-30T11:09:10+03:00

The paper proposes a variant of implementing a partial synthesizer with a small frequency step and preserving a sufficient level of phase noise in the X-band frequency range, which provides high frequency stability and low phase noise by combining 3 methods. A brief review of common methods for constructing frequency synthesizers, such as phase-locked loop, digital signal synthesis (DDS), dielectric resonator oscillator, is given. Their advantages and disadvantages were used and taken into account in the development of a new method for constructing a frequency synthesizer. The article compares the characteristics of the phase-locked loop frequency synthesizer on the ADF5355 chip with the developed method. The proposed method, which includes all 3 proposed methods, is presented in the form of a functional circuit containing two phase-locked loops and one DDS. The first PLL contains a dielectric resonator oscillator with an output signal of 8 GHz and a working frequency bandwidth of 1 kHz with a minimum phase noise equal to –132.85 dBc/Hz at a 1 kHz offset. The low-noise DDS signal is fed to the second phase-locked loop. The output signal is in the range of 9-9.5 GHz with a phase noise of –98.32 dBc/Hz at a 10 kHz offset.

Programmed Module for Transportation of Humanitarian Medical Cargo by Drones in Extreme Conditions

2025-09-30T10:38:45+03:00

The structure and functions of various drones (UAVs) for medical purposes were observed, their general analysis was done. The main goal and objectives of the work were outlined: to develop the structure of a container for UAVs for medical purposes, and to clarify the possibility of its transportation to programmatically defined object. These were done basing on the analysis of prototypes – versions of previously developed UAV modules for providing various types of medical assistance in extreme conditions. Different versions of modules for medical UAVs were characterized. The need to deliver a container to a specific object or person requires the creation of appropriate software to automate this process; developed software was based on face recognition technology – convolutional neural network. The work on the development of a container structure for medical purposes was disclosed in details. Its technical task was to present a version of medical container for multirotor type UAV with vertical take-off and landing. The medical container itself was made in the form of a streamlined cylinder to improve its aerodynamic characteristics, which reduces the value of air resistance during flight and increases the overall maneuverability and stability of the structures. Developed medical UAV was equipped with modules for diagnosing person's condition and providing them with appropriate medical care. Developed UAV can be used repeatedly, and modular systems of medical equipment and first aid supplies can be parachuted over the location of potentially injured person. The modules can be conditionally divided into several groups: diagnostic, resuscitation, a subgroup for transporting biomaterials, devices for detecting possible chemical contamination, etc. In process of the work, there were subdivided and developed several such modules (groups of devices) in the newly developed UAV. The data about the practical application of work results were given: possibility of introducing a UAV with a first aid container into the activities of emergency services (ambulance, other emergency services); use for detecting of chemical substances, environmental pollutants using chemo-specific detectors in conditions of potential danger to human life and health; provision of medical care in extreme conditions; delivery of necessary medications (vaccines, medical preparations) or performing the function of a courier for delivering biological samples to the nearest laboratory.

Calibration of Pressure Measurement Channels in Wind Tunnels

2025-09-29T20:36:53+03:00

The article deals with estimating errors in the measurement systems during experimental tests of unmanned aerial vehicles in the wind tunnel. The instrumentation and technique used for calibrating pressure parameters are described. Features of the measurement in the wind tunnel are characterized. The main sources of pressure measurement are listed. The features of measuring pressure in the wind tunnel are discussed. The structural diagram of the calibrating channel of pressure measurement. The transformation function of the measuring channel was proposed. The approach for estimating relative errors is represented. The estimation of high-velocity head measurement errors has been carried out. Relative and absolute errors of velocity head were estimated. Errors in measuring airflow speed have been estimated. It is shown that the velocity of the airflow is the result of an indirect nonlinear measurement. The obtained results can be useful for testing unmanned aerial vehicles of different types. They can also be applied to the measurement of the pressure in various experimental equipment.

Automated Technological Design оf Nanoscale Transistors

2025-09-29T19:57:01+03:00

The article is devoted to the automation of technological preparation of modern high-frequency and energy-efficient bipolar transistors with nanoscale depths of impurity implantation into a nanoconductor substrate. At the stage of mathematical modeling of technological operations of multilayer casting, the known theoretical and empirical, developed by the authors of the article, high-temperature dependences of doping parameters, distributions of depths of boundary distances of emitter and collector junctions, which, as a result, determine the thickness of the electrically neutral base region of the transistor, are taken into account. Mathematical models of technological parameters of surface and volume concentrations of impurities, which cause degeneration and repeated inversion of the conductivity types of the initial crystalline substrate, are proposed. The maximum possible values of the tincture and solution of acceptor and donor impurities, which increase the gain coefficients and reduce the power consumption of bipolar nanotransistors, are determined. The drift components of the base and collector currents, which are caused by the internal electric field of the inhomogeneous base, are taken into account. The temperature and time dependences of technological doping operations are found, which primarily determine the creation of bipolar transistors with a base thickness from 100 nm to 10 nm. The values of the limiting concentrations of impurities in semiconductor structures are established. Examples are considered that confirm the effectiveness of the proposed methods for automated design of bipolar nanotransistors. In the future, it is planned to develop generalized algorithms for multi-level hierarchical modeling of transistor nanoelectronics components.

Optimizing Drone Coverage in Agriculture: an Overview and New Approaches

2025-09-29T19:24:39+03:00

This article investigates the problem of trajectory optimization for unmanned aerial vehicles during multispectral imaging of agricultural lands within the framework of precision agriculture concepts. The main problems related to complex field geometry, presence of natural and artificial obstacles, as well as limited battery capacity of drones are considered. A new hybrid route optimization method is proposed that integrates the ant colony optimization algorithm for global planning of zone traversal sequence with the binary gridding method for detailed local replanning within complex areas and obstacle avoidance. A key feature of the method is an adaptive mission recovery mechanism that allows the drone to dynamically return to the charging station, save mission state, and automatically continue operation from the last uncovered area. Simulation and comparative analysis results demonstrate that the developed approach significantly reduces total traveled route length and optimizes mission execution time compared to traditional methods, confirming its effectiveness for increasing autonomy and productivity of agricultural unmanned aerial vehicles.

Decentralized Local-priority Communication Protocol for Small Unmanned Aerial Vehicle Swarms

2025-09-29T14:53:31+03:00

The paper proposes a decentralized communication protocol for small swarms of unmanned aerial vehicles that provides prioritized access to the control channel with limited radio resources. The approach is based on local priority selection, agent slot mapping with seat rotation for long-term fairness, and probabilistic sparsity within. This combination manages the load in a mathematical expectation, reduces the probability of collisions, and ensures low latency delivery of priority control messages without a central dispatcher. The simulation results for a swarm of 12 unmanned aerial vehicles demonstrate an increase in usable throughput, median delay at the level of one epoch, and collision rate at the level of the baseline approach with a significantly higher number of successful transmissions.

A Classification Method for Optical Coherence Tomography Images Based on a Structure-oriented Adaptive Neural Network Architecture

2025-09-29T14:29:17+03:00

The method of optical coherence tomography image classification for automated diagnosis of diabetic retinopathy and diabetic macular edema is proposed in the article. An innovative adaptive multi-task deep neural network is created. It simultaneously solves the problems of pathology classification and structural feature reconstruction. The neural network uses the pre-trained EfficientNetB7 model as an encoder for efficient extraction of high-level features. The structural feature learning branch (decoder) is responsible for restoring spatial information. It increases the resolution of feature maps to the original size of 224x224 pixels with a gradual decrease in the number of filters and the use of Batch Normalization to stabilize learning. The classification branch combines semantic and structural features. It uses the channel attention mechanism for dynamic weighting of informative channels. Dropout and Batch Normalization layers are used to prevent overtraining in the classification branch. The model is optimized using a multi-task loss function. It consists of a modified loss function for classification (with class weights to balance data imbalance) and a root-mean-square error for structural loss. Training is performed using the Adam optimizer and the EarlyStopping, ModelCheckpoint, and ReduceLROnPlateau callbacks. The experiment was conducted on the OCT Image Classification dataset. Data augmentation (horizontal reflections) was performed to increase the number of images. High accuracy rates and cost functions were obtained as a result of training. The multi-task method enables the encoder to learn details and boundaries of the retina through Canny edge reconstruction. It contributes to improved classification and provides a powerful internal regularization mechanism, increasing the generalization ability of the model.

Re-uploading Data in Tensor Network

2025-09-29T13:51:06+03:00

In this paper, we present an approach for enhancing quantum tensor networks through the method of data re-uploading. The proposed framework integrates multiple layers of classical data encoding into tensor network architectures, thereby improving their approximation capacity and reducing the impact of barren plateaus in training. The model construction relies on tree tensor networks combined with RX, RZ, and RY rotational gates and CNOT entanglement, while optimization is performed using differential evolution as a gradient-free algorithm. Experimental evaluation was carried out on the iris and wine datasets, comparing baseline tensor networks with architectures incorporating one to three re-uploading layers. The results demonstrate a consistent reduction in training and test loss, with accuracy, recall, and precision reaching 100% on the iris dataset for three layers and improvements of up to 40% in prediction quality on the wine dataset. These findings confirm that data re-uploading significantly enhances the performance and expressiveness of tensor network-based quantum models.

A Formal Language for the Analysis of Graph Models and its Software Implementation

2025-09-29T13:07:43+03:00

The purpose of this paper is to develop a specialized language for processing graph data and its software implementation. The proposed solution ensures versatility, usability, and efficiency, enabling the execution of both basic graph operations and more complex procedures. The tool supports classical graph algorithms, including shortest path search, graph traversal, and minimum spanning tree construction, as well as applications in modeling and analyzing message transition processes and processing graph-based representations of textual data. A common drawback of many existing graph analysis tools—often implemented as libraries of general-purpose programming languages—is their limited usability. This limitation arises from the fact that the description of graph analysis procedures relies on data structure operations defined in terms of these general-purpose languages, which complicates perception and reduces the clarity of the mathematical methods being implemented. Developing a specialized domain-specific language based on high-level abstractions can address these shortcomings. Such a language will provide a formalized description of methods for analyzing and processing graph models, improving their comprehensibility and accessibility to users. Its software implementation will deliver ready-to-use solutions for executing graph analysis methods. Composing such methods will facilitate solving a wide range of tasks, including the analysis of natural language messages and the study of information publication and dissemination processes in online environments.

Intelligent System for Diagnosing Vestibular Schwannoma

2025-09-29T11:46:56+03:00

This scientific work is devoted to the development of an intelligent system for the diagnosis of vestibular schwannoma. A new approach to the texture analysis of magnetic resonance images of vestibular schwannoma is proposed in order to determine the assessment of tumor growth. The use of this approach will prevent the risks of tumor progression and timely determine the need for surgical intervention. Several classes of texture descriptors were used in the study, including: first-order statistics (intensity histograms), gray level co-occurrence matrix, gray level run length matrix, gray level size zone matrix, gray level dependency matrix, as well as wavelet-transformed features. The complex use of these descriptors made it possible to formalize the internal microstructure of the tumor and implement an effective model for predicting its growth.

Flight Safety Issues During Aircraft Landing Approach

2025-07-01T12:56:08+03:00

This paper examines the psychophysiological stress experienced by flight crews during a standard landing approach, a phase considered one of the most critical in aviation operations. Elevated levels of cognitive and emotional strain can significantly influence pilot performance, including situational awareness, decision-making accuracy, and reaction time. The study emphasizes the pivotal role of timely and precise avionics diagnostics in mitigating risks associated with equipment failure or misinterpretation of instrument data. In this context, a method for calculating the reliability of critical onboard instruments during landing is proposed. The approach integrates both technical reliability factors and operational stressors to provide a comprehensive evaluation framework. The findings aim to support enhanced safety protocols and inform the development of more resilient avionics systems.

Mathematical Model of Gyroscope with Contactless Suspended Rotor and Three-component Accelerometer for Solving Problems of Autonomous Autonomous Navigation

2025-07-01T11:58:23+03:00

The problem of creating a mathematical model of a free three-degree gyroscope with a contactless suspended rotor and a three-component accelerometer for solving problems of high-precision autonomous navigation is solved. The problem under consideration includes determination of kinematic relations of the gyro device and solution of a direct problem for finding navigation parameters for a stationary base. The suspension system of these gyroscopes is practically indifferent to the environment in which the gyro operates, but in order to reduce braking moments, the gyro rotor is placed in a vacuum chamber. The given structure of vector relationships between coordinate systems and model parameters allows us to write down several groups of equations regarding the angular positions of the gyro rotor and its angular velocities.

Classification of Sentinel-2 Imagery Using Rayleigh Distribution Modeling

2025-07-01T11:23:15+03:00

Nowadays land cover classification from satellite imagery is one of most actual and important problems in remote sensing. Multispectral satellite images such as Sentinel-2 images provide high-resolution imagery in different spectral bands, enabling detailed distinguishing of surface objects. This study presents a method of multispectral satellite image classification based on Rayleigh distribution, maximum likelihood method and likelihood functions. It was considered three land cover classes, such as “Water”, “Vegetation”, and “Buildings”, applying three spectral bands (Red spectral band, Green spectral band and Blue spectral band). Proposed classification procedure includes modeling spectral distributions with the Rayleigh probability distribution. The Rayleigh distribution parameters for each class and each spectral band are estimated from training data via the proposed formula. The ESA SNAP software is applied for image processing. Maximum likelihood method is applied for classification procedure. In remote sensing this method is used to classify pixels in satellite imagery into different classes. This method is based on assigning each pixel to the class, for which has the highest probability of belonging. It was described the methodology, including data preparation using the ESA SNAP software and data analysis in Microsoft Excel. The mathematical formulation of the Rayleigh distribution and the mathematical algorithm of calculation of likelihood functions for each class and for each spectral band have been considered. Results include fitted Rayleigh distribution parameters for each class and for each spectral band, classification maps, calculation of likelihood functions and classification result. The classification result depends on which class the maximum likelihood function corresponds to. An example has been considered where the class “Vegetation” is determined using the maximum likelihood method and Rayleigh distribution. The proposed approach can be applied for land-cover classification, ecological monitoring, agriculture and geological tasks.

A Method for Biometric Coding of Speech Signals Based on Adaptive Empirical Wavelet Transform

2025-07-01T10:53:41+03:00

In this research, a biometric speech coding method is developed where empirical wavelet transform is used to extract biometric features of speech signals for voice identification of the speaker. This method differs from existing methods because it uses a set of adaptive bandpass Meyer wavelet filters and Hilbert spectral analysis to determine the instantaneous amplitudes and frequencies of internal empirical modes. This makes it possible to use multiscale wavelet analysis for biometric coding of speech signals based on an adaptive empirical wavelet transform, which increases the efficiency of spectral analysis by 1.2 times or 14 % by separating high-frequency speech oscillations into their low-frequency components, namely internal empirical modes. Also, a biometric method for encoding speech signals based on mel-frequency cepstral coefficients has been improved, which uses the basic principles of adaptive spectral analysis using an empirical wavelet transform, which also significantly improves the separation of the Fourier spectrum into adaptive bands of the corresponding formant frequencies of the speech signal.

Prediction of Moving Targets and Adaptive Avoidance in Hybrid PSO-MPC for a Swarm of UAV’s

2025-07-01T10:24:11+03:00

The paper proposes a hybrid approach to the safe control of a multicopter swarm in the presence of two moving obstacles based on a combination of the particle swarm algorithm and model predictive control. The first stage of the algorithm is a global search for new target positions of subgroup centers using particle swarm algorithm based on predicted data, which allows the front subgroup to smoothly climb and avoid the danger zone. The second stage is the local adjustment of the movement of each vehicle within the model predictive control, taking into account dynamic constraints, which ensures accurate adherence to the calculated targets and prevents formation disruption. Simulation experiments demonstrate that the developed algorithm ensures coordinated maneuvers of all subgroups, timely avoidance of both moving threats, and return to the original formation without sudden jumps in altitude or chaotic behavior.

Accuracy Research for Non-orthogonal Configuration of Inertial Sensors

2025-07-01T09:57:28+03:00

This article deals with accuracy research of the non-orthogonal configuration of inertial sensors based on Allan variance. The influence of changes in the measurement range of the inertial module on the Allan variance was assessed. Based on an analysis of the results of the Allan variance assessment, a procedure for choosing multi-axis MEMS sensors with identical characteristics to create an inertial non-orthogonal measuring instrument is proposed. An example of compiling a data processing algorithm for an inertial measuring instrument with a non-orthogonal arrangement of sensitivity axes based on an assembly of 3-axis MEMS sensors is given. The simulation results for numerical estimates are represented. Improvement of the accuracy of the non-orthogonal inertial measuring instruments using the Allan variance is shown.

Security System for Office Premises with Use of Modern Information Technologies

2025-06-30T12:47:08+03:00

The paper analyzes the office premises and determines the required set of functions and structure of the security system, which includes a video surveillance subsystem, an access control subsystem, a burglar alarm subsystem, and a backup power supply subsystem. A detailed scheme of placement and interaction of components is proposed, and algorithms for the system operation are presented. The integration of the backup power supply subsystem with renewable energy sources ensures autonomous operation in the event of power grid failures.Office security is critical for protecting employees, property, and sensitive information by integrating physical and cybersecurity measures. Modern systems use advanced information technologies, such as remote, high-resolution video surveillance with autonomous analytics and biometric-based access control, to monitor, detect, and respond to threats efficiently. This technological integration enables real-time oversight and immediate action, significantly enhancing overall office protection.

Air Raid Alert Mesh Network System: Key Provision

2025-06-30T12:01:24+03:00

In the face of modern hybrid threats and infrastructure vulnerabilities, the timely and secure dissemination of air raid alerts is vital for civilian safety. Traditional centralized alert systems are susceptible to disruption, making decentralized wireless alternatives increasingly relevant. This paper presents a secure key provisioning framework for a decentralized air raid alert system built on LoRa-based mesh networking. Each node is equipped with a cryptographic identity stored in a hardware secure element (ATECC608A), enabling message authentication, signature verification, and node revocation without centralized control. The proposed system ensures that only trusted nodes can initiate or relay alert signals, effectively preventing spoofing, replay attacks, and unauthorized activations. A series of real-world tests and simulations demonstrates that the framework introduces minimal latency while significantly enhancing system resilience and trustworthiness. The results confirm the feasibility of a scalable, tamper-resistant alert network capable of operating under degraded or hostile conditions.

Diagnosis of Vestibular Schwannoma Based on Intelligent MRI Image Processing

2025-06-30T10:08:35+03:00

The study identified the main clinical and diagnostic features of the disease, reviewed modern diagnostic methods for schwannoma, including magnetic resonance imaging and computed tomography, as well as the role of clinical examination, history and laboratory tests, analyzed available open data and proposed the concept of combining medical images with molecular indicators to build more effective diagnostic models based on semantic segmentation. Diagnostic and prognostic biomarkers were summarized, including TNF-α, CD68, CD163, IL-6, CCR2 and others, which may increase the accuracy of predicting the course of the disease.

A Comprehensive Benchmark of Collaborative Filtering Methods on Implicit Feedback Datasets

2025-06-30T09:46:57+03:00

Collaborative filtering is a foundational technique in modern recommender systems, especially when dealing with implicit feedback signals such as clicks, purchases, or listening behavior. Despite the abundance of сollaborative filtering models, including classical, probabilistic, and neural approaches, there is a lack of standardized, large-scale evaluations across diverse datasets. This study presents a comprehensive empirical benchmark of 13 сollaborative filtering algorithms encompassing matrix factorization, pairwise ranking, variational and non-variational autoencoders, graph-based neural models, and probabilistic methods. Using four representative implicit feedback datasets from different domains, we evaluate models under a unified experimental protocol using ranking-based metrics (MAP@10, NDCG@10, Precision@10, Recall@10, MRR), while also reporting training efficiency. Our results reveal that neural architectures such as NeuMF, VAECF, and LightGCN offer strong performance in dense and moderately sparse scenarios, but may face scalability constraints on larger datasets. Simpler models like EASEᴿ and BPR often achieve a favorable balance between performance and efficiency. This benchmark offers actionable insights into the trade-offs of modern сollaborative filtering methods and guides future research in implicit recommender systems.

Parametric Optimization of the Hierarchical Fuzzy Model of Control with Transfer of Fuzzy Values of Intermediate Data

2025-06-28T19:04:39+03:00

The subject of the study is the intellectualization of the technological process of controlling complex objects in order to intellectualize and replace the labor of a human operator. In conditions that are difficult to describe by mathematical methods due to incompleteness and uncertainty, a hybrid neuro-fuzzy model with a hierarchical structure is used to control the process. The aim of the article is to study and develop a learning algorithm for the Mamdani→Sugeno model with the transfer of fuzzy intermediate data between hierarchical levels, implemented by an adaptive neural network. To ensure the accuracy of real-time forecasting, an algorithm for parametric adaptation to operating conditions with the adjustment of the parameters of antecedents and consequences at two levels has been defined. When studying the methods of data transfer between levels, fuzzy logic and artificial neural networks methods, the gradient descent method, Mamdani and Takagi–Sugeno–Kang algorithms, etc. were used. The study confirms the possibility of using hybrid models to intellectualize the process of controlling complex objects. The scientific innovation of the obtained results is the construction of a neural network of a hierarchical control system and the development of a learning algorithm for the transfer of fuzzy intermediate variables with parametric model adaptation based on the gradient descent algorithm.

Orthophotomosaicing Framework for Thermal and Multispectral Images Collected with a UAV for Intelligent Systems

2025-06-28T18:38:11+03:00

In this paper, a framework for orthophotomosaicing of multispectral and thermal images collected by unmanned aerial vehicles is presented. The proposed framework is based on a two-stage data preprocessing and mosaicing orthophotographic restoration of images captured with a route-planned unmanned aerial vehicle collection. The super-resolution and image restoration step is handled via a two-pathway U-net image restoration artificial neural network. The framework simplifies the process and makes the collected data less sensitive to noise via image restoration and upscaling steps. The framework was tested on visible, multispectral and thermal images and provides 3.5% and 5.34% improvements in peak signal-to-noise ratio for multispectral and thermal orthophotomosaics.

Multi-agent Control of UAVs Using Deep Reinforcement Learning

2025-06-28T16:07:27+03:00

This paper presents a novel control framework for managing a group of unmanned aerial vehicles using multi-agent deep reinforcement learning. The approach leverages actor–critic architectures, centralized training with decentralized execution, and shared experience replay to enable autonomous coordination in dynamic environments. Simulation results confirm improved tracking accuracy, reduced collision rates, and increased coverage efficiency. The study also compares the proposed system against baseline methods and outlines future work for real-world adaptation. The novelty lies in applying multi-agent deep reinforcement learning to a continuous unmanned aerial vehicle control task in cluttered environments with limited sensing.

Optimizing Kubernetes Autoscaling with Artificial Intelligence

2025-06-28T10:58:05+03:00

This study explores how to improve Kubernetes auto-scaling using artificial intelligence based forecasting. The authors emphasize the limitations of traditional, reactive auto-scaling methods that lag behind rapid changes in demand and propose a proactive approach that predicts future resource requirements. The paper presents a framework for integrating artificial intelligence based predictions into the Kubernetes ecosystem to improve operational efficiency and resource utilization. To address the main challenges, the authors focus on improving workload forecasting and mitigating the impact of random fluctuations in Kubernetes performance. To address this issue, they use time-series forecasting models combined with data preprocessing techniques to predict future CPU utilization and thus inform scaling decisions before peaks or troughs in demand occur. The results show that artificial intelligence based forecasting can significantly improve scaling accuracy, reduce latency, and optimize resource utilization in Kubernetes environments. Time-series models are developed and evaluated using real CPU utilization data from a Kubernetes cluster, including RNN, LSTM, and CNN-GRU. The study also explores new architectures such as Fourier Analysis Network and Kolmogorov–Arnold Network and their integration with the transformer model. In general, the proposed approach aims to improve resource efficiency and application reliability in Kubernetes through proactive automatic scaling.