The subject of this study focuses on the interpretability of fault-detection frameworks in solar cyber-physical systems. The rapid expansion of renewable energy networks generates massive volumes of continuous sensor data, requiring sophisticated monitoring strategies to prevent safety hazards such as photovoltaic fires. Nevertheless, dominant deep learning models function as opaque black boxes, providing no transparent reasoning for critical safety decisions. The goal of this work is to enhance the interpretability of anomaly detection in cyber-physical environments by formulating a rule-based production knowledge base straight from continuous sensor readings. Key tasks include developing an adaptive discretisation method, identifying minimal feature subsets, and reconciling contradictory patterns using an integral class support score. The applied methods fuses rough set theory (RST) with an innovative weighted entropy-density discretization (WEDD) algorithm. This combined pipeline fine-tunes thresholds using a dual metric of information entropy and local probability density, utilizing kernel density estimation to position cut points within natural data valleys. Deterministic rules are derived from the lower approximation, whereas probabilistic rules are generated from the boundary region. The results illustrate the substantial effectiveness of the suggested approach. Tested on a simulated SCADA dataset designed for fire hazard identification, the framework attains an overall accuracy of 96.2% and a macro F1-score of 0.960. Significantly, it delivers 100% accuracy for deterministic rules and a 98.0% recall rate for the vital Fire Hazard category. Comparative evaluations on the Iris and Wine datasets yield competitive results, achieving 93.3% and 87.6% accuracy, respectively, when compared with Decision Tree and Naive Bayes models. The generated knowledge base is a compact JSON artifact, making it well-suited for edge devices with limited computational resources. In conclusion, this research establishes the WEDD-RST approach as a rigorous approach for transforming raw sensor data into fully auditable IF-THEN rules with explicit confidence scores, offering a highly reliable solution for automated safety monitoring in cyber-physical environments.

This paper presents a novel hypersector-based method with Fuzzy Learning Vector Quantization (FLVQ) for the real-time classification of wind turbine blade defects using data acquired by unmanned aerial vehicles (UAVs). Unlike conventional prototype-based FLVQ approaches that rely on Euclidean distance in the feature space, the proposed method models each defect class as a hypersector on an n-dimensional hypersphere, where class boundaries are defined by angular similarity and fuzzy membership transitions. This geometric reinterpretation of FLVQ constitutes the core innovation of the study, enabling improved class separability, robustness to noise, and enhanced interpretability under uncertain operating conditions. Feature vectors extracted via the pre-trained SqueezeNet convolutional network are normalized onto the hypersphere, forming compact directional clusters that serve as the geometric foundation of the FLVQ classifier. A fuzzy softmax membership function and an adaptive prototype-updating mechanism are introduced to handle class overlap and improve learning stability. Experimental validation on a custom dataset of 900 UAV-acquired images achieved 95% classification accuracy on test data and 98.3% on an independent dataset, with an average F1-score of 0.91. Comparative analysis with the classical FLVQ prototype demonstrated superior performance and noise robustness. Owing to its low computational complexity and transparent geometric decision structure, the developed model is well-suited for real-time deployment on UAV embedded systems. Furthermore, the proposed hypersector FLVQ framework is generic and can be extended to other renewable-energy diagnostic tasks, including solar and hydropower asset monitoring, contributing to enhanced energy security and sustainability.

Automated visual diagnostics of renewable energy assets require efficient transformation of inspection video streams into compact and informative representations. Raw inspection videos of wind turbine blades contain a large number of redundant frames, while surface defects such as cracks, erosion, and corrosion occur only locally. Direct pixel-level processing of such data is computationally inefficient for streaming applications. This paper presents a SqueezeNet-based software module for processing inspection video streams and extracting compact four-dimensional class-confidence vectors. The proposed approach performs frame sampling, standardized image preprocessing, and convolutional feature extraction, followed by a lightweight classification head that produces 4D probability-based feature representations. The focus is placed on feature compactness and temporal stability rather than classification accuracy alone. Experimental results obtained on real inspection video data demonstrate clear class separability in the representation space and high temporal consistency, with an average cosine similarity of 0.988 between consecutive frames. The proposed representation is well-suited for integration into intelligent diagnostic systems operating in real-time or near-real-time conditions.

The article develops an intelligent decision support system for personnel evacuation and replacement in the context of hybrid threats, integrating deep learning and multi-criteria optimisation methods. The main task is to predict personnel’s functional state and determine priorities for their evacuation or replacement in real time, while accounting for simultaneous cybernetic, physical, and information-psychological threats. The study uses a convolutional neural network (CNN) module for classifying personnel status and a long short-term memory (LSTM) module for predicting the probability of functionality loss. The study’s novelty lies in the complex integration of these technologies for human resource management in emergencies, as well as in the development of an adaptive model for dynamic threat assessment and personnel status. An important feature is the use of multi-criteria methods for decision-making on evacuation and replacement with limited resources. The experimental results show an increase in the predictive accuracy of personnel’s functional state to 90% and effective resource optimisation, ensuring loss minimisation in crises. The research contribution is to create a sustainable, scalable system that improves decision-making effectiveness in the face of hybrid threats and reduces risks to national security.

The subject of research is the architectural advancement of inspection systems for large-scale solar power plants. As global solar infrastructure expands, reliance on manual or offline analytical methods creates significant operational bottlenecks. The goal of research is to improve the operational utility of unmanned aerial vehicle (UAV)-based photovoltaic module inspection by developing a cyber-physical system (CPS) architecture. It integrates onboard deep learning, edge nodes, cloud analytics, and Supervisory Control and Data Acquisition (SCADA)-aware decision-making into a single coordinated workflow. The tasks of research: 1) formalise a multi-tiered CPS architecture (UAV-edge-cloud) and define interfaces for data, geo-tags, and alarms; 2) develop and validate an onboard thermographic detection pipeline with palette-aware fusion; and 3) integrate detection results with a SCADA-aware logic layer for hazard inference and fire risk mitigation. The methods of research: Computer vision and deep learning (YOLOv11) are used for onboard defect segmentation. Model ensembling via a late-fusion strategy for M2 and M3 thermal palettes mitigates domain shift. RTK-supported spatial clustering algorithms ensure precise geo-indexing and deduplication, and deterministic Boolean logic assesses fire risks based on bypass diode states. The results obtained with five-fold cross-validation shows the proposed architecture significantly outperforms single-modality baselines. The onboard YOLOv11 model achieved a macro mAP@0.5 of 0.91 and 0.90 for M2 and M3 palettes, respectively. The late-fusion ensemble elevated mAP@0.5 for cracks to 0.96 and delamination to 0.95. It reduced end-to-end per-frame processing latency from 4.235 s to 2.858 s. Field validation demonstrated an error of 0.71 defects per inspected string compared to manual counts. Sensitivity analysis highlighted that a 10 m flight altitude provides an optimal balance, yielding 93% precision and 90% recall. Conclusions: Treating UAV inspection as an integrated cyber-physical service improves defect detection. This offers a scalable, real-time solution for preventive maintenance and automated fire-risk mitigation in renewable energy.

Water monitoring data are often irregular, incomplete, and lack reliable quality labels, which limits supervised methods. This paper proposes a station-specific, label-free one-class pipeline for multivariate anomaly detection in water-quality monitoring. Isolation Forest is trained separately for each station after completeness-based feature filtering, median imputation, and robust scaling. Detected events are explained using robust deviations, reporting the top contributing indicators for each anomaly. Experiments on 2023–2024 data from river reaches, reservoirs, and an estuary demonstrate data-driven anomaly counts across stations and interpretable indicator-level profiles suitable for expert review and operational validation.

Independent restructuring of the architecture of multicomputer systems during their operation is a complex task, since such systems are distributed. One of the tasks in this restructuring is to change the architecture of system centers. That is, the system can be rebuilt without changes in its center. But the specifics of the tasks of systems for detecting malicious software and computer attacks require such an organization of systems that it is difficult for attackers to understand their behavior. Therefore, the current task considered in the work is the development of rules for ensuring the restructuring of system centers according to different types of architecture. The aim of the work is to develop criteria for evaluating potential options for centralization in the architecture of multicomputer systems with traps and decoys. To ensure such an assessment, the work analyzed known solutions and established the insufficiency of mathematical support for organizing the restructuring of system centers during their operation. Taking into account the specifics of the tasks for such systems, no parameters were determined that could be taken into account for the formation of the restructuring of system centers. The analyzed works establish the main types of centralization used in the architecture of systems: centralized, partially centralized, partially decentralized, decentralized. However, algorithms and methods for the transition of systems from one type to another in the process of their functioning are not provided. Subject. The work defines characteristic properties that can be used when synthesizing systems. They determine the number of potential variants of the system architecture to which it will switch at the next step when making a decision on restructuring the architecture. With an increase in the number of characteristic properties, the number of possible variants will increase. When approving the variants for the transition, it was necessary to evaluate them taking into account the previous experience of the systems' functioning. To evaluate potential centralization variants in the architecture of systems, evaluation criteria were developed. A feature of the evaluation criteria is that according to them, it is possible to take into account the experience of using the centralization variant in the case of repetition and evaluate the prepared variants that are offered for the first time. That is, the evaluation criteria include the previous experience of the functioning of multi-computer systems. This experience made it possible to evaluate the repeated option based on the results of its previous use. This made it possible to diversify the choice of system centers. Methods. The work developed an objective function for evaluating the next centralization option in the system architecture. The objective function takes into account four evaluation criteria for operational efficiency, stability, integrity and security. All these criteria are focused on evaluating potential options for system centers. New mathematical models were developed for the criteria for operational efficiency, stability, integrity and security in relation to the system center, which, unlike the known mathematical models for evaluating system centers for selecting the next options for centralization, are presented in analytical expressions that take into account the features of the types of centralization in the system architecture, indicators of operational efficiency, stability, integrity and security in relation to the system center and allow forming on their basis an objective function for evaluating options for centralization in systems, the feature of which is the hiding of components with the system center from detection by attackers. Results. The work analyzed the results of an experiment conducted with a prototype of the system. The convergence of the experimental results and the results obtained by the theoretical method has been established. Conclusion. The study introduces mathematical models for evaluating system centers based on operational efficiency, stability, integrity, and security criteria. Unlike existing models, these are presented as analytical expressions that account for various centralization types within system architectures. The models enable the creation of objective functions to evaluate centralization options, emphasizing the concealment of system center components from attackers. Experimental results with a system prototype confirm the theoretical models' validity, showing minimal deviations in function graphs. Significant deviations in specific time intervals are addressed to achieve optimal centralization options.

This article develops a method for analysing sensor data to prevent cyberattacks using a modified LSTM network. This method development is based on the fact that in the context of the rapid increase in sensor devices used in critical infrastructure, it is becoming an urgent task to ensure these systems’ security from various types of attacks, such as data forgery, man-in-the-middle attacks, and denial of service. The method is based on predicting normal system behaviour using a modified LSTM network, which allows for effective prediction of sensor data because the F1 score = 0.90, as well as on analysing anomalies detected through residual values, which makes the method highly sensitive to changes in data. The main result is high accuracy of attack detection (precision = 0.92), achieved through a hybrid approach combining prediction with statistical deviation analysis. During the computational experiment, the developed method demonstrated real-time efficiency with minimal computational costs, providing accuracy up to 92% and recall up to 89%, which is confirmed by high AUC = 0.94 values. These results show that the developed method is effectively protecting critical infrastructure facilities with limited computing resources, which is especially important for cyber police.

The Subject of this research is the development of an intelligent, integrated system for the early detection and causal analysis of fire hazards in large-scale solar power plants (SPPs). It addresses the critical shortcomings of conventional monitoring methods, which often lack the necessary integration, speed, and diagnostic depth to reliably prevent catastrophic failures resulting from photovoltaic (PV) module defects. The goal of this study is to design, develop, and validate a comprehensive, multi-modal framework that fully automates the monitoring workflow, from data acquisition to actionable decision-making. The proposed system aims to significantly enhance plant safety by providing reliable, cause-differentiated alerts, which in turn optimizes maintenance strategies, minimizes downtime, and improves the overall economic viability of solar energy infrastructure. The Methods employed involve a synergistic architecture that combines an Unmanned Aerial Vehicle (UAV) equipped with high-resolution RGB and radiometric infrared cameras for rapid imaging, supplemented by dedicated Internet of Things (IoT) temperature sensors on PV module bypass diodes for critical component verification. A custom-trained YOLOv8 deep learning model performs automated defect detection from the captured imagery. The system’s intellectual core is a novel logical inference engine based on a Disjunctive Normal Form (DNF) equation. This formal logic model intelligently fuses four key binary features, namely, primary defect cause (damage vs. soiling), visual evidence, thermal anomaly severity, and bypass diode functional status, to produce a definitive and context-aware fire risk assessment. The entire workflow is managed and visualized using a SCADA TRACE MODE platform for centralized control and automated alerting. The study successfully validated the performance and logical integrity of the integrated system through a series of high-fidelity, scenario-based simulations. These simulations rigorously confirmed the capability of the DNF logic to accurately and reliably identify all predefined fire hazards. This included not only obvious faults but also "stealthy," damage-induced hotspots where the primary safety mechanism (the bypass diode) had failed. Concurrently, the system correctly classified mitigated risks to prevent false alarms, demonstrating its diagnostic precision. This capability allows the system to reliably differentiate between true emergencies requiring immediate module replacement and less critical issues, such as soiling that merely necessitates cleaning. The projected increase in diagnostic accuracy for identifying critical, fire-prone defects over conventional, single-modality methods is up to 40%, providing a quantitative measure of enhanced safety and reliability. Furthermore, the proposed system is projected to reduce the false-positive alarm rate by over 75% compared with IR-only automated systems. In conclusion, this study establishes a powerful new paradigm for proactive SPP safety management. The intelligent fusion of UAV and IoT sensing, AI-driven analytics, and a formal logical framework provides a robust and reliable solution for fire risk mitigation, enabling a highly efficient, condition-based maintenance strategy and significantly enhancing the safety, reliability, and performance of modern solar power infrastructure.

This research presents the development of an intelligent controller for the helicopter turboshaft engines gas temperature, aimed at compensating for the measuring sensor’s inertial delays and optimizing transient processes. The aim is to compensate for inertial delays τ ≈ 0.025 seconds and optimize transient processes. The method is based on a double summation circuit with a channel selector, comparing signals from a thermocouple and a gas-generator rotor speed sensor, and an adaptive observer based on Pade approximation and Taylor series expansion provides a prediction of the state at t + τ. The intelligent control law includes a proportional-integral-differential structure with the coefficients γi correction via gradient descent. To refine the delay estimate, a two-layer fully connected multilayer perceptron with a SmoothReLU activation function is implemented trained on flight test data was implemented, which reduced τ to 0.016 seconds (–36 %). This module allows to approximate nonlinear relations between input features and the delay value, which ensures the control signal’s timely correction and the system’s adaptation to changing operating conditions. Modeling of the system in the Matlab Simulink environment demonstrated a significant improvement in the transient process characteristics: overshoot was reduced from 8.0 to 1.5 %, and the mode establishment time was reduced from 4.2 to 3.3 seconds. The neural network module testing showed high predicting accuracy (99.537 % with losses of 0.511 %), confirmed by the determination coefficient R2 = 0.9717. The neural network use made it possible to reduce the delay value to 0.016 seconds, which corresponds to an improvement of 36 % compared to traditional methods. The obtained results indicate a proposed technique’s high potential for improving the helicopter turboshaft engines automatic control system’s dynamic accuracy and stability.

The detection of polymorphic viruses poses significant challenges due to their ability to modify their structure while retaining malicious functionality. This paper presents a hybrid multi-agent system (MAS) architecture designed for the detection and classification of polymorphic viruses in computer networks. The system consists of specialized agents-Analysis, Propagation Rate, Detection, Classification, and Decision- making agents-that collaborate to analyze threats, model virus spread, detect infections using diverse methods, classify viruses by complexity levels using fuzzy logic, and apply appropriate mitigation strategies. The effectiveness of the proposed MAS was evaluated using additive and multiplicative performance models, achieving an average efficiency of 95.35% across varying complexity levels of polymorphic viruses. Experimental results demonstrate the system's high detection and classification rates, particularly for lower-complexity threats, and validate its potential for deployment in cybersecurity operations. Future improvements will focus on enhancing agent adaptability and detection precision against highly complex polymorphic malware.

This research is a research continuation in the PID controller development and modernization field based on neural networks. Through the pseudo-linear component with amplitude suppression and another with phase advance incorporation, embedded within the nonlinear PID controller structure, previously employed effectively as a mechanism for regulating the helicopter turboshaft engines free turbine rotor speed with a linear electronic regulator, has undergone modernization. The nonlinear PID controller is implemented as a dynamic neural network that directly transmits data, comprising neurons with a radial basis activation function in the initial layer and adalines neurons with a linear activation function in the subsequent layer. In alignment with the aforementioned, this neural network undergoes modification. The pseudo-linear link with amplitude suppression is integrated as an additional layer comprising linear neurons with a sigmoid activation function. Meanwhile, the pseudo-linear link with phase advance is integrated as a parallel layer featuring linear neurons with a sigmoid activation function. Test research findings demonstrate accuracy levels of up to 0.998 in addressing the intricate dynamic systems (demonstrated by helicopter turboshaft engines free turbine rotor speed) controlling parameters task.

This article presents a method for researching processes in automatic control systems based on the operator approach for modelling the control object and the controller. Within the method framework, a system of equations has been developed that describes the relations between the control error, the reference and control action, the output coordinate and the controller and the control object operators. The traditional PI controller modification, including a switching function for adaptation to operating conditions, allows for the system’s effective control in real time. The controller optimization algorithm is based on a functional expression with weighting coefficients that take into account control errors and the control action. To train the neural network through implementing the proposed method, a multilayer architecture was used, including nonlinear activation functions and a dynamic training rate, which ensure high accuracy and accelerated convergence. The TV3-117 turboshaft engine was chosen as the research object, which allows the method to be demonstrated in practical applications in aviation technology. The experimental results showed a significant improvement in control characteristics, including a reduction in the gas-generator rotor speed parameter transient time to ≈1, which is two times faster than the traditional method, where the transient process reaches ≈0.5. The model achieved a maximum accuracy of 0.993 with 160 training epochs, minimizing the error function to 0.005. In comparison with similar approaches, the proposed method demonstrated better results in accuracy and training speed, which was confirmed by a reduction in the number of iterations by 1.36 times and an improvement in the mean square error by 1.86–6.02 times.

Unmanned aerial vehicles (UAVs) equipped with advanced sensors have opened up new opportunities for and other critical components. However, reliable defect detection ficient to enhance of an ensemble of YOLO-based deep learning models that integrate both visible and thermal channels. We propose an ensemble approach that integrates a general-purpose YOLOv8 model with a specialized thermal model, using a sophisticated bounding box fusion algorithm to combine their predictions. Our experiments show this approach achieves a mean Average Precision (mAP@.5) of 0.93 and an F 1-score of 0.90, outperforming a standalone YOLOv8 model, which scored an mAP@.5 of 0.91. These findings demonstrate that combining multiple YOLO architectures with fused multispectral data provides a more reliable solution, defects.

An effective neural network system for monitoring sensors in helicopter turboshaft engines has been developed based on a hybrid architecture combining LSTM and GRU. This system enables sequential data processing while ensuring high accuracy in anomaly detection. Using recurrent layers (LSTM/GRU) is critical for dependencies among data time series analysis and identification, facilitating key information retention from previous states. Modules such as SensorFailClean and SensorFailNorm implement adaptive discretization and quantisation techniques, enhancing the data input quality and contributing to more accurate predictions. The developed system demonstrated anomaly detection accuracy at 99.327% after 200 training epochs, with a reduction in loss from 2.5 to 0.5%, indicating stability in anomaly processing. A training algorithm incorporating temporal regularization and a combined optimization method (SGD with RMSProp) accelerated neural network convergence, reducing the training time to 4 min and 13 s while achieving an accuracy of 0.993. Comparisons with alternative methods indicate superior performance for the proposed approach across key metrics, including accuracy at 0.993 compared to 0.981 and 0.982. Computational experiments confirmed the presence of the highly correlated sensor and demonstrated the method’s effectiveness in fault detection, highlighting the system’s capability to minimize omissions.

Modern imaging systems produce a great volume of image data. In many practical situations, it is necessary to compress them for faster transferring or more efficient storage. Then, a compression has to be applied. If images are noisy, lossless compression is almost useless, and lossy compression is characterized by a specific noise filtering effect that depends on the image, noise, and coder properties. Here, we considered a modern HEIF coder applied to grayscale (component) images of different complexity corrupted by additive white Gaussian noise. It has recently been shown that an optimal operation point (OOP) might exist in this case. Note that the OOP is a value of quality factor where the compressed image quality (according to a used quality metric) is the closest to the corresponding noise-free image. The lossy compression of noisy images leads to both noise reduction and distortions introduced into the information component, thus, a compromise should be found between the compressed image quality and compression ratio attained. The OOP is one possible compromise, if it exists, for a given noisy image. However, it has also recently been demonstrated that the compressed image quality can be significantly improved if post-filtering is applied under the condition that the quality factor is slightly larger than the one corresponding to the OOP. Therefore, we considered the efficiency of post-filtering where a block-matching 3-dimensional (BM3D) filter was applied. It was shown that the positive effect of such post-filtering could reach a few dB in terms of the PSNR and PSNR-HVS-M metrics. The largest benefits took place for simple structure images and a high intensity of noise. It was also demonstrated that the filter parameters have to be adapted to the properties of residual noise that become more non-Gaussian if the compression ratio increases. Practical recommendations on the use of compression parameters and post-filtering are given.

The paper considers the application of the Wang-Mendel neuro-fuzzy network to the problem of classifying defects in wind turbine blades. The proposed model combines the principles of fuzzy logic and artificial neural networks, which ensures high accuracy even under limited sample conditions. The network was trained based on input data containing the probabilities of the presence of various types of damage. The model achieved an accuracy of 94% and demonstrated the ability to effectively adapt the parameters of membership functions and weight coefficients. The results of the study confirm the effectiveness of using hybrid neuro-fuzzy systems in the technical diagnostics of energy equipment.