Given the significance of the electricity industry in today’s world, the role of virtual power plants (VPPs) in electricity markets must be closely examined as a key issue. These power plants, utilizing a wide range of distributed generation resources, operate as a single entity in energy markets. This study investigates innovative pricing strategies for VPPs participating in both reserve and energy markets under market price uncertainty. VPPs aggregate diverse distributed energy resources and require robust optimization tools to effectively navigate dynamic market conditions. A genetic algorithm–based model is proposed to develop profit-maximizing bidding strategies considering technical constraints, load demands, and price volatility. The model is applied to an 18-bus test system across three scenarios: without uncertainty, with partial uncertainty, and with full uncertainty in market prices and generation. Results show that the proposed method improves total profit by up to 4.5% under uncertainty scenarios compared to baseline and effectively shifts operational decisions to maximize profitability, particularly during low-price hours. This demonstrates the model’s robustness and strategic value in complex electricity markets.
In modern power systems, increasing flexibility has become an essential aspect of supporting the integration of renewable energy resources (RERs). The rapid fluctuations in power output from RERs and the reduced reliance on conventional thermal units necessitate flexible energy resources and demand-side solutions. This study introduces a control framework that utilizes thermostatically controlled loads (TCLs) represented by electric water heaters (EWHs) to support the power system operation in delivering both primary and secondary frequency responses, especially for situations that lack adequate ramping capability from conventional resources. At the operator level, load frequency regulation is performed using a model predictive control (MPC) framework enhanced with integral action to control the frequency in a two-area power system. The MPC is responsible for adjusting the thermal power plant outputs and providing the reference control signals for demand response agents (DRAs), which are responsible for load aggregation, monitoring, and control. DRAs based on state-space statistical learning techniques develop an aggregate model within a linear system framework, which is used in a linear quadratic regulator (LQR) to optimally manage device operation and achieve the reference control signals. The developed approach can transform thousands of individual binary devices into a compact, fixed-dimensional state-space model. Therefore, the computational burden remains independent of the number of underlying devices, proving its viability for large-scale power system implementations. This study considers 10,000 heterogeneous EWHs distributed in each control area. MATLAB simulations demonstrate that TCLs can effectively contribute to primary and secondary frequency control, offering a scalable solution for integrating TCLs into future grid operations and improving grid flexibility.
Reliable spectrum sensing is a key requirement for cognitive radio (CR) networks; however, its performance is significantly degraded by fading, noise uncertainty, and channel estimation errors. This paper proposes a hybrid gated recurrent unit–long short‐term memory–convolutional neural network (GRU–LSTM–CNN) framework for cooperative spectrum sensing (CSS) that integrates spatial–spectral feature extraction with long‐ and short‐term temporal learning to improve primary user detection. The proposed framework incorporates a reliability‐aware cooperative fusion mechanism in which secondary users (SUs) transmit soft sensing reports together with predictive uncertainty, enabling adaptive weighting based on signal quality and decision reliability. Extensive simulations under additive white Gaussian noise (AWGN), Rayleigh, and Rician fading channels demonstrate that the proposed method consistently outperforms conventional energy detection (ED), matched filter (MF), long short‐term memory (LSTM), convolutional neural network (CNN), and LSTM–CNN models. The proposed framework achieves a probability of detection (Pd) approaching 1 at a signal‐to‐noise ratio of −4 dB, provides improvement in bit error rate (BER) performance over conventional methods, significantly reduces the probability of false alarm (Pfa), achieves the lowest power spectral density (PSD), and maintains robust sensing performance under 15% and 25% channel estimation errors. These results demonstrate that the proposed GRU–LSTM–CNN framework delivers accurate, robust, and reliable CSS, making it a promising solution for next‐generation internet of things (IoT)‐enabled and sixth‐generation (6G) wireless communication systems.
The paper describes the design, modeling, and experimental validation of a wideband, high‐gain antenna array for sub‐6 GHz 5G applications, which uses a unique guitar‐shaped radiating element combined with complementary split‐ring resonators (CSRRs) and an M‐shaped metasurface ground. Three array topologies (1 × 2, 2 × 2, and 3 × 2 corporate‐feed networks) are studied to improve bandwidth (BW) and directional radiation performance. The inclusion of three CSRRs in each element allows for triple‐band operation, which includes crucial sub‐6 GHz bands and sections of the X‐band, while the M‐shaped metasurface enhances impedance matching, suppresses surface waves, and increases gain. CST Microwave Studio 2025 is used to optimize the antenna arrays, and PCB prototyping is used to construct them. Peak gains of 6.64 dBi (1 × 2), 6.98 dBi (2 × 2), and 8.908 dBi (3 × 2) are shown in the measured findings, which were acquired using S‐parameter characterization with a vector network analyzer and radiation pattern measurements in an anechoic chamber. The observed impedance BWs range from 488 MHz to 2.88 GHz. For contemporary sub‐6 GHz 5G communication systems, the suggested design ensures excellent signal quality and minimal distortion by providing high efficiency and broad performance together with steady group delay throughout the working bands.
The increasing integration of renewable energy sources into hybrid microgrids has significantly improved the sustainability of power systems. However, the intermittent nature of solar and wind generation makes real-time energy management a major challenge. Conventional energy management strategies often exhibit limited adaptability to rapidly changing operating conditions, resulting in inefficient power dispatch, increased fuel consumption, and accelerated battery degradation. To address these limitations, this study proposes an intelligent energy management framework based on the integration of a gated recurrent unit (GRU) deep learning model with model predictive control (MPC) for a grid-connected hybrid microgrid supplying residential buildings in Maroua, Cameroon. The GRU model predicts load demand and renewable energy generation using historical and real-time data, while the MPC optimally schedules power flows by considering system constraints, battery state of charge, renewable energy availability, and load requirements. The proposed GRU–MPC strategy was implemented and validated in MATLAB/Simulink using real operational data from the study area. Simulation results demonstrate that the proposed approach achieves a 30% reduction in fuel consumption, a 15% improvement in overall system efficiency, a 20% reduction in battery stress, and a forecasting accuracy of 99.8% compared with conventional energy management approaches. In addition, the proposed framework improves renewable energy utilization, maintains the battery state of charge around 65%, ensures continuous power supply, and reduces the operating energy cost to approximately 25 USD under the investigated scenarios. These results confirm that the GRU–MPC strategy provides an effective, robust, and scalable solution for intelligent energy management in modern hybrid renewable energy microgrids.
This study introduces PolarDoG–SlimVGG, a novel framework for the automated classification of complex quadruple power quality disturbances (PQDs). The proposed method combines PolarDoG, a feature extraction technique based on polar coordinate transformation and difference of Gaussians (DoG) edge detection, with a lightweight convolutional neural network, slimmed-down VGGNet (SlimVGG), optimized for efficient and accurate classification. Comprehensive evaluations were conducted on 13 PQD classes under varying noise levels (10–40 dB). The framework achieved state-of-the-art performance, with an accuracy of 0.997 under noise-free conditions, and demonstrated high robustness, maintaining accuracies of 0.992 at 20 dB and 0.936 at 10 dB. The SlimVGG model exhibits fast convergence, low overfitting, and computational efficiency, while the PolarDoG feature extractor outperforms conventional signal processing techniques in retaining discriminative features under noisy conditions. Compared with recent approaches, the proposed framework demonstrates superior performance in handling multidisturbance scenarios and enhanced noise tolerance. These findings underscore the potential of PolarDoG–SlimVGG as a practical and robust solution for real-time PQD monitoring and analysis in smart grid applications.
As the integration of photovoltaic systems and battery energy storage devices into distribution networks increases, issues such as node voltage violations and higher network losses have emerged. To address these challenges, this paper proposes an optimal allocation strategy for energy storage systems in distribution networks based on an improved particle swarm optimization algorithm, namely, EP-MOPSO. This strategy incorporates a nondominated sorting mechanism and grid adaptation to handle multiobjective optimization problems and employs an explosive particle strategy to enhance global exploration capability, suppress premature convergence, and maintain population diversity. An optimization model is established with the objectives of minimizing the economic cost of the energy storage configuration and optimizing its technical performance. The effectiveness of the proposed method is validated through simulations on the IEEE 33-bus distribution system. Simulation results show that the total network loss is reduced by 1250.90 kW, the voltage profile is significantly improved, and the maximum voltage improvement outperforms other intelligent algorithms by over 37%. On the IEEE 69-bus distribution system, the total network loss is reduced by 1457.66 kW, and the proposed strategy maintains the voltage stability of each line within a safe range. Compared with other intelligent algorithms, the maximum voltage improvement advantage exceeds 25%, while effectively reducing the installation cost of energy storage systems, thereby contributing to enhanced voltage stability of the distribution network.
In modern agriculture, detecting crop conditions and pest infestations is crucial for improving yield and quality. Automated crop detection is a key component of intelligent agriculture. Therefore, this study proposes an improved multiobject analysis framework for crop monitoring and pest prevention, aiming to increase the accuracy and real-time performance of agricultural health monitoring. The improved YOLO structure, AgriFocus-YOLO, is built on the YOLO11 architecture, improving the spatial pyramid pooling-fast (SPPF) module by integrating the large separable kernel attention (LSKA) mechanism for better multiscale feature extraction. To address the challenges of feature representation in agricultural tasks, we introduce the multiscale dilated fusion attention (MDFA) module, which uses parallel multidilation convolution and channel-spatial dual attention to enhance object detection, classification, and segmentation. The MDFA module effectively integrates local and global features, improving recognition in complex scenes. Additionally, we design a collaborative architecture—the efficient shared convolution head (ESCH) and the task dynamic alignment detection head (TDADH)—to optimize task synergy by dynamically adjusting classification and regression gradients. The WCIoU bounding box loss, ABFL focus loss, and LSDW-CE classification loss are employed to increase the detection, classification, and segmentation performance. The experimental results show that the proposed method achieves a 94.9% mean average precision (mAP) in detection, a 97.7% binary classification accuracy, and an 85.6% mAP in segmentation. These results demonstrate the effectiveness of the method for crop health monitoring.
Reconfigurable intelligent surfaces (RISs) are a promising technology for improving mobile communication performance due to their ability to shape the propagation environment. In this paper, an RIS-assisted beamforming framework is proposed with low computational complexity. This framework is designed for a downlink multiuser-multiple input single output (MU-MISO) system where the base station uses discrete Fourier transform (DFT) beamforming codebook. RIS phase shifts are designed using a simple coherent alignment strategy to avoid computationally expensive joint optimization. In order to prepare the beam selection, a convolutional neural network (CNN) is utilized, and holistic Monte Carlo simulations are handled for eight users to assess the system performance in terms of beam selection accuracy, received power, spectral efficiency, outage probability, and RIS gain. Top-K achieves a very high prediction performance, and the framework provides gains between −6 and +1 dB according to the channel conditions. Furthermore, the proposed system achieves zero outage probability under the considered SNR scenario and simulation settings despite low computational complexity level compared with optimization-based RIS systems. The results confirmed that RIS assistance enhanced beamforming performance under favorable channel conditions and achieved measurable power gain while still being a low-complexity system. The analysis of the performance of the system reveals a trade-off between the RIS improvement and interuser inference, so the system presents a practical and scalable solution for intelligent beam management in future RIS-assisted wireless communication systems. Furthermore, the results of the multiuser system reveal that there is a trade-off between RIS-induced gains and the system robustness. This conclusion provides practical insights into the deployment of RIS-assisted beamforming systems for 6G networks.
Acute lymphoblastic leukemia (ALL) is a critical hematological malignancy that requires rapid and accurate diagnosis. Most existing automated detection methods rely on segmented datasets and exhibit limited generalization to clinical microscopy images. This study aims to develop and evaluate LEUKENet, an explainable deep ensemble framework for robust, accurate leukemia detection across both segmented and nonsegmented images. It integrates three convolutional neural network architectures—DenseNet-121, MobileNetV2, and VGG16—into an ensemble classifier. The model was trained on both versions of the ALL-IDB2 dataset and subsequently tested on two independent datasets (ALL-IDB1 and blood cell cancer) to assess cross-dataset generalization. Cell-level explainability was achieved using Grad-CAM to visualize discriminative regions within leukemic cells. Statistical robustness was evaluated using 1000 bootstrap resamples. Furthermore, the complete system was deployed as a lightweight web-based clinical decision support tool. LEUKENet achieved 98.3% accuracy and 99.22% AUC on the segmented ALL-IDB2 dataset. It generalized effectively to unseen datasets, achieving 93.52% and 90.04% accuracy on nonsegmented ALL-IDB1 and blood cell cancer images, respectively. Bootstrap analysis confirmed statistically significant improvements (p<0.001 against chance; nonoverlapping 95% CIs between configurations) across key performance metrics. Our model demonstrates benchmark performance, strong cross-dataset generalization, and statistical reliability, along with a practical diagnostic support solution for clinical and resource-limited healthcare environments.
The electrocardiogram (ECG) is considered the gold standard for cardiac arrhythmia diagnosis; however, its manual interpretation is slow, subjective, and prone to human error. Deep learning (DL) techniques, especially convolutional neural networks (CNNs), have demonstrated significant performance in automatic classification and detection of biomedical signals. Coleeg is an open-source framework that was originally designed for EEG signal classification; however, its modular design makes it easy to extend for other types of signals. This paper utilizes the Coleeg framework to investigate cross-subject classification using a one-dimensional CNN (1D-CNN) for three arrhythmia databases, which are MIT-BIH, INCART, and PTB-XL, and it presents the features introduced in Coleeg Versions 6.0 and 7.0, which are related to this research work. The main problem encountered in the classification is the extreme imbalance in the databases, where normal beats dominate database classes, which reduces the classification performance for other beat types. To mitigate the data imbalance, the normal beat class count was reduced, and class weights were implemented during training. Data augmentation using the Synthetic Minority Over-sampling Technique (SMOTE) was evaluated, but it did not yield a significant or consistent improvement in the evaluation metrics. Because of the fluctuation in validation loss, early stopping was used to obtain the best classification results. The model evaluation was performed in a systematic way for the three databases, where they were resampled to 250 Hz, and the subjects were divided into 5 groups for a fivefold evaluation. One group is used for validation, one for testing, and the remaining three are for training. Each database evaluation was repeated 5 times for 1-, 2-, and 3-s time windows. The details of the numerical values for the minimum, mean, maximum, and sample standard deviation of recall, precision, F1-score, and accuracy are presented in this paper.
Future wireless networks must accommodate massive connectivity under stringent spectral- and energy-efficiency requirements, thereby intensifying the demand for robust power-domain nonorthogonal multiple access (NOMA). The performance of NOMA critically depends on reliable successive interference cancellation (SIC), whose effectiveness deteriorates under imperfect channel state information (CSI) due to estimation errors, mobility, and hardware nonlinearities. To mitigate these challenges, this paper proposes an adaptive particle swarm optimization (PSO)-based power allocation framework that enhances SIC robustness by dynamically learning optimal power coefficients in the presence of CSI uncertainty. The problem is reformulated as a constrained multiobjective optimization task that incorporates decoding-order stability, interference propagation effects, user fairness, and rate performance. The proposed adaptive PSO integrates CSI uncertainty directly into its update dynamics through error-aware adjustment of inertia and acceleration parameters. Extensive numerical evaluations conducted under realistic imperfect-CSI assumptions show that the proposed method significantly improves fairness, sum-rate stability, and SIC error resilience, achieving up to 98% successful SIC operations in medium-to-high signal-to-noise ratio (SNR) regimes. These findings demonstrate the practicality of lightweight, AI-driven optimization techniques for improving the reliability of NOMA-enabled 5G-and-beyond wireless systems.
With the increasing penetration of photovoltaic (PV) installations and their inherent intermittency due to weather variability, advanced energy control strategies have become essential for modern microgrid operation. In this context, reliable short-term forecasting and optimal decision-making are critical to ensure stable and efficient energy management. This work proposes a predictive energy management framework based on a parallel deep learning architecture combining long short-term memory (LSTM) and gated recurrent unit (GRU) networks, whose hyperparameters are optimized using a multiobjective particle swarm optimization (MOPSO) algorithm for accurate PV power forecasting. The developed forecasting model is explicitly integrated into a model predictive control (MPC) scheme for real-time microgrid energy management. In this MPC framework, the LSTM–GRU–MOPSO model provides multi-step-ahead PV power predictions that serve as key inputs to the optimization problem solved at each control interval (5 min). The MPC then determines optimal power dispatch decisions, including energy storage charging/discharging and grid interaction, while respecting system constraints and minimizing operational cost and power imbalance. The proposed approach is validated using a real PV dataset collected in Maroua, Far North Cameroon, under daily, weekly, and monthly prediction horizons. Results demonstrate that the hybrid forecasting model achieves high accuracy across daily, weekly, and monthly horizons, with R2 values of 0.9984, 0.9988, and 0.9934, and RMSE values of 0.2096, 0.2034, and 0.3501, respectively. Comparative analysis shows that the proposed LSTM–GRU–MOPSO model outperforms conventional methods such as LSTM, GRU, support vector machine (SVM), and feedforward neural networks (FFNN), as well as hybrid architectures including LSTM–CNN and LSTM–RNN. The integration with MPC further enhances system reliability by enabling predictive, constraint-aware, and adaptive microgrid energy management.
This paper proposes an unified predictor-based sliding-mode control (SMC) architecture tolerant to sensor bias for nonlinear input-delayed systems. The approach integrates predictor feedback for delay compensation and an unknown-input observer (UIO) for online bias reconstruction. A complete input-to-state stability (ISS) proof under bounded prediction and observation errors is provided, along with practical tuning guidelines. A detailed case study on a nonisothermal CSTR demonstrates a 70% performance improvement over baseline methods under 12-s input delay and +6°C sensor bias, with effective bias estimation within 30 s of fault occurrence. Comparative analysis with baseline and RL-based FTC methods confirms the method’s robustness and practical applicability.
This paper presents a new design method of a functional observer for a bilinear system class with unknown input and constant time delay present in both state and input vectors. The proposed observer is developed in time and frequency domains. The time domain solution is based on Lyapunov-Krasovskii stability theory, which is transformed on linear matrix inequalities (LMIs) conditions. It provides an estimation of a functional state vector. The frequency domain method is synthetized by adopting a coprime factorization approach applied on time-domain results. The proposed method is tested on a numerical example, which proves its efficiency.
Reliable operation of modern transmission networks requires not only postfault detection but also predictive identification of instability before fault escalation. This paper presents a probabilistic SIL-based framework for real-time transmission line monitoring and predictive fault detection using receiving-end phasor measurements. Unlike conventional transient-driven methods, the proposed approach formulates an adaptive SIL stability boundary incorporating voltage deviation and reactive power imbalance and transforms SIL exceedance into a hazard-based probabilistic risk metric. Monte Carlo simulations (N = 200 per loading level) validated exponential instability growth beyond the adaptive threshold, yielding a calibrated risk constant, k = 0.1746, with strong regression agreement, R2 = 0.9793. Implementation on a 220 kV, 200 km transmission corridor and dynamic validation on an IEEE 9-bus multimachine system demonstrated nonlinear escalation of fault probability beyond 110% SIL. The framework achieved 94.7% detection accuracy across 570 test disturbances and identified high-risk conditions up to 50 ms earlier than an adaptive wavelet transform benchmark, while maintaining lower computational complexity. A two-dimensional SIL-reactive compensation risk map further enables proactive operator actions such as dynamic VAR support and preventive load shedding. The results establish SIL deviation as a computationally efficient early-warning indicator for transmission-line instability, enabling predictive fault risk assessment and scalable deployment in wide-area monitoring, and modern grid protection systems.
Security operations centers (SOCs) face escalating challenges from alert fatigue and a critical skills gap, leading to delayed incident response times. Addressing these persistent issues calls for innovative automation and decision-support approaches. Large language models (LLMs) present a transformative opportunity to automate complex workflows and augment the capabilities of human analysts. This survey provides the first comprehensive and structured analysis of LLM integration into SOC operations. We systematically map LLM applications to the functions of the NIST Cybersecurity Framework (CSF) and use the MITRE ATT&CK framework as an analytical lens to evaluate their granular threat detection capabilities. By synthesizing findings from 216 studies, we provide a structured overview of current methodologies, identify key trade-offs between LLM architectures, and highlight significant gaps in research, particularly in the NIST “Recover” function. This survey offers researchers and SOC managers a clear research roadmap and actionable insights for leveraging LLMs to build more resilient and intelligent security operations.
This paper proposes an AI-based light fidelity (Li-Fi) and millimeter-wave (mmWave) architecture to hastily mitigate most of the challenges in vehicle-to-vehicle (V2V) communication systems. Through traffic prediction based on machine learning technologies, the system enables proactive resource allocation, as well as intelligent frequency management in the unpredictable vehicle environment. Traffic forecast is realized by means of a simple linear extrapolation, based on the knowledge of drifting vehicles' speed and position; optimization techniques based on Q-learning algorithms allow for adaptive frequency assignment such as to reduce interference and increasing spectral efficiency. Simulation results show the desirable enhancements in communication quality, average SINR values around 25-30 dB, and communication delays less than 3-5 ms at medium traffic scenario. Hybrid Li-Fi/mmWave infrastructure: up to 10 Gbps (Li-Fi zones) and reliability reaching up to 100 m (mmWave zones). These observations suggest the promising direction of AI-powered hybrid communication systems (HCSs) to improve vehicle networks for more intelligent and safe autonomous transportation landscape.
To overcome the limitations of traditional intrusion detection methods in dealing with high-dimensional sparse features, multiclass attack classification, and model robustness, this paper presents a fused multibranch intrusion detection model (FMB-IDM). The proposed framework combines three complementary deep learning components. First, an MLP branch equipped with a Squeeze-and-Excitation attention module is used to enhance feature representation by adaptively reweighting channel-wise information. Second, a TabTransformer-based branch is introduced to model complex dependencies among structured input features through self-attention. Third, a CNN-BiLSTM branch is employed to capture both local feature patterns and longer-range sequential relationships, where the CNN extracts local representations, and the two-layer bidirectional LSTM learns contextual dependencies. To make better use of the information learned by each branch, an attention-based fusion strategy is adopted to combine their outputs adaptively, which improves the model’s ability to distinguish different types of intrusion behaviors. Experiments conducted on the NSL-KDD, UNSW-NB15, and CIC-IDS2017 datasets demonstrate that the proposed model achieves strong and consistent performance in both binary and multiclass intrusion detection tasks. In particular, on CIC-IDS2017, FMB-IDM reaches an accuracy of 97.56% and a weighted F1-score of 0.98. In addition, the model maintains good inference efficiency, with an average latency of about 0.03 ms per sample under the experimental hardware configuration.
Hydrophobicity is a critical property influencing the performance and reliability of high-voltage insulators, particularly nonceramic types, by mitigating surface contamination and reducing the risk of flashover. This paper presents a comprehensive review of hydrophobicity assessment techniques, spanning traditional laboratory-based techniques, alternative experimental approaches, and modern artificial intelligence (AI)–assisted systems. Standard methods such as the contact angle, surface tension, and spray techniques are evaluated for their accuracy and limitations in field deployment. Alternative methods, including salt fog and dynamic drop tests, offer enhanced insights but remain constrained by laboratory requirements. The review emphasizes the rapid evolution of AI-driven techniques, ranging from image processing and pattern recognition to deep learning models such as convolutional neural networks (CNNs) and vision transformers (ViTs), which enable scalable, automated, and objective surface condition categorization. Despite their promise, these methods face challenges related to image quality, environmental variability, and computational demands. To address these, the paper proposes a structured framework for developing a robust machine learning model tailored for hydrophobicity classification. The framework outlines key stages including data acquisition, model architecture, training strategies, evaluation metrics, and deployment considerations, with a focus on real-time, field-ready applications. This review not only synthesizes the current state of the art but also provides a roadmap for future research and standardization efforts aimed at integrating AI-based solutions into practical insulator condition monitoring.
