This study proposes a dual-stream BiLSTM framework for household load forecasting that integrates time-series dynamics with histogram-based daily shape features. Unlike existing models relying on weather or external data, the proposed method extracts intrinsic load-shape information directly from normalized daily curves. A multihead attention module fuses temporal and shape representations, enabling adaptive weighting of informative dimensions. Experiments on three real-world datasets show consistent improvements over the baseline BiLSTM, with up to 30.12%, 24.27%, and 19.03% reductions in MAE, RMSE, and SMAPE, respectively. The results highlight the framework’s robustness and efficiency for fine-grained load forecasting without external inputs.
Textual resources are among the most valuable sources of information in cognitive neuroscience (CN) for understanding and investigating brain activity and cognitive processes. Extracting and constructing knowledge graphs (KGs) from these texts can facilitate medical research by providing deeper insights into neurological diseases and brain function. In recent years, the use of large language models (LLMs) in natural language processing (NLP) has become increasingly widespread, significantly enhancing the extraction of meaningful information from large volumes of text. This study proposes a novel approach for constructing and evaluating a specialized knowledge graph, termed the cognitive neuroscience knowledge graph (CNKG), from scientific publications in the field of CN by leveraging the capabilities of GPT-4. During the construction process, GPT-4 is employed to extract relationships among predefined CN concepts from scientific texts. The resulting graph is then refined to maximize its accuracy and representativeness. Finally, the quality and performance of the CNKG are assessed using GPT-4-based evaluation procedures. The evaluation yielded an accuracy score of 0.936. In addition, link prediction analysis demonstrated that the proposed KG possesses satisfactory quality. Furthermore, complex network metrics obtained using Gephi, particularly the average clustering coefficient (0.419541) and graph diameter (13), provided additional evidence supporting the validity of the constructed graph. The CNKG has the potential to support a variety of downstream applications, including semantic query answering, recommendation systems, and research aimed at the diagnosis and treatment of neurological diseases and disorders. Moreover, it may contribute to improving the quality of research services within the field of cognitive neuroscience. Overall, the proposed approach offers considerable potential to enhance the efficiency and accuracy of cognitive neuroscience literature analysis, thereby opening new avenues for scientific investigation and discovery.
The workforce scheduling and routing problem (WSRP) involves assigning tasks across multiple locations while accounting for varying travel times, service durations, time windows, and skill requirements in a wide range of industries, from healthcare to telecommunications. This paper presents a mixed-integer programming model for the WSRP that balances the trade-off between cost and customer satisfaction using a score-generation function and subsequently evaluates the trade-off between solution quality and computation time for several algorithms on well-known datasets. We demonstrate that our model effectively balances cost, service-level agreement satisfaction, and task priorities while providing high-quality solutions in a timely manner. Observing that the best-performing algorithm varies across problem instances, we propose an algorithm selection approach based on a gradient boosting classifier that achieves 89\% of the performance gap between the virtual best solver and the single best solver.
This study proposes a novel variable-gain mechanism with a minimal number of tuning parameters to enhance the performance of conventional backstepping controllers for nonlinear systems while avoiding singularity and peaking phenomena. The proposed approach is simple, computationally efficient, and well suited for real-time implementation without imposing a significant computational burden. Its effectiveness is validated through real-time experiments conducted using a dSPACE DS1104 controller board and a 7.5-kW induction motor (IM). Simulation results demonstrate that the proposed controller outperforms the conventional backstepping controller. Robustness analyses under variations in stator resistance, load inertia, and viscous friction coefficient reveal substantial reductions in the integral squared error (ISE), from 42.78 to 0.73, 67.4 to 1.38, and 3.65 to 0.53, respectively. Experimental results further confirm that the proposed method achieves superior tracking performance and enhanced robustness against disturbances compared with existing methods reported in the literature.
Plant leaf disease detection (PLDD) is a growing active research area with burgeoning practical applications across various sectors such as agricultural monitoring, food security, and environmental conservation. Accurate segmentation and classification of plant leaf diseases remains a key challenge in the field of plant leaf disease prediction. The challenge demands automated methods for the plant disease identification because it needs to develop better crop management systems, which will boost agricultural production. In this article, we provide a systematic review of various machine learning (ML) and deep learning (DL) methods extensively used for PLDD. The review strategy follows a formal protocol, involving structured search, screening, and analysis of studies published between 2020 and 2024. We have proposed a taxonomy of PLDD methods that will be useful for experts and researchers working in this exciting research area. The review thoroughly examines techniques for both segmentation and classification of the PLDD workflow. In addition, we examine several public and private datasets that are accessible to study plant diseases and highlight their significance in developing accurate diagnostic models. The paper also presented multiple performance assessment criteria that researchers can use to evaluate PLDD methods at present and in the future. The study also discusses the current challenges in plant leaf disease classification and offers essential insights about upcoming developments and potential enhancements. The research findings from this study provide essential knowledge that helps experts and researchers to develop automated systems to detect and classify plant leaf diseases effectively.
Solar panels are becoming very essential in providing sustainable energy but they are usually affected by defects on the surface like dust, snow, bird droppings, physical damages and electrical faults which interfere with their performance. These faults must be identified accurately and in a timely manner to enhance energy efficiency, lower the maintenance cost, and supplement the traditional manual methods of inspection which are labor-intensive, time-consuming and subject to human errors in judgment. The most common methods, such as traditional CNNs and hybrid architectures tend to be less accurate, less explainable and cannot be properly evaluated to be deployed in real time. In order to overcome these issues, we proposed SwinDeiTViT an ensemble model that fuses Swin-Tiny and DeiT-Small vision transformers using a soft voting layer. The framework uses sophisticated preprocessing and data augmentation in order to increase feature visibility and model generalization. SwinDeiTViT reaches an overall accuracy of 99.31%, cross-validation accuracy of 98.41%, F1-scores of approximately 1.0 and a Cohen’s Kappa of 0.9915, with low inference latency that is compatible with edge deployment on a varied set of 826 solar panel images expanded to 4320 samples. Also, Grad-CAM visualizations provide transparent interpretability by highlighting regions contributing to predictions.The proposed framework can provide a functional, operational solution to real-time monitoring of solar panels, which is useful to enhance manual inspection, and be more robust, reliable, and explainable compared to a baseline model.
Few-shot image classification benefits from data augmentation, yet most existing methods operate in pixel space with limited control over spectral semantics. We introduce a lightweight, frequency-guided augmentation strategy based on Variational Mode Decomposition (VMD). Our method constructs an offline, per-class ModeBank by decomposing downsampled luminance patches and retaining midband modes that encode class-specific texture patterns. During episodic training, VMD is never executed online: instead, for each support image, a same-class midband mode is selected and blended using PSNR-targeted scaling with a luminance energy cap, ensuring perceptual consistency. The augmentation is fast, reproducible, class-consistent, and integrates seamlessly into standard metric-based pipelines without modifying the backbone or training procedure. We evaluate our approach on multiple benchmarks, including CIFAR-FS, FC100, Tiny-ImageNet-200, and DTD under 5-way 1-shot and 5-shot settings, using Conv-4 and ResNet-12 backbones. Our results show consistent and statistically significant gains over strong ProtoNet baselines, with effect sizes ranging from small to large and most improvements observed in the challenging 1-shot regime. Ablation studies confirm the importance of targeted frequency selection, query-side consistency, energy/PSNR control, and adaptive mixing schedules. Our method demonstrates that structured, class-aligned frequency perturbations offer a principled and practical complement to spatial augmentations in few-shot learning.
Deep Neural Networks (DNNs) have achieved remarkable success across diverse machine learning applications, yet designing effective architectures remains a laborious, expert-driven process. Neural Architecture Search (NAS) was introduced to automate this process, with Evolutionary NAS (ENAS) emerging as one of the most effective and widely adopted NAS paradigms. This survey provides a comprehensive and systematic review of 164 ENAS studies published between 2020 and 2024, categorized according to the specific evolutionary algorithm employed as the search strategy. Unlike prior surveys—which either treat evolutionary methods at a high level or focus on general NAS pipelines—this study is, to the best of our knowledge, the first to perform an in-depth, algorithm-level breakdown of ENAS research, covering evolutionary strategies (ES), genetic algorithms (GA) and their derivatives, Particle Swarm Optimization (PSO), Differential Evolution (DE), Quantum-Inspired Evolutionary Algorithms (QIEA), and other heuristics. Our quantitative analysis reveals that ES is the dominant paradigm (45.7% of studies), followed by GA (29.9%), with annual publication counts growing 6.6-fold from 2020 to 2024—reflecting a rapidly expanding field. Key findings include: (i) surrogate-assisted methods have grown substantially in 2023--2024, significantly reducing search cost; (ii) GA-based multiobjective approaches (NSGA-II, NSGA-III) dominate Pareto-optimal architecture search; (iii) hybridization of evolutionary search with gradient descent and predictor networks is an accelerating trend; and (iv) image classification on CIFAR-10/100 and NAS-Bench-201 benchmarks accounts for the majority of evaluations, highlighting opportunities in underexplored domains. This survey also provides a critical comparison of encoding strategies (fixed-length vs. variable-length) and an analysis of search space complexity across reviewed methods. We believe this work serves as a practical guide for researchers seeking to understand, select, and advance evolutionary strategies for automated neural architecture design.
Vehicular networks support intelligent transportation through vehicle-to-roadside Units (V2R) and vehicle-to-vehicle (V2V) communication but face challenges from dynamic topologies, limited RSU coverage, and bandwidth scarcity, which impact service delivery and revenue. RDA-ITU addresses these challenges by integrating V2R and V2V paradigms to maximize RSU revenue, enhance service availability, and improve system efficiency. It dynamically allocates services based on real-time network conditions and vehicle mobility, leveraging V2V relays to optimize both RSU-direct and cooperative communication. Through extensive simulations, RDA-ITU significantly outperforms four baselines: RBSM, VVMM-U, VVMM-LW, and VVMM-MA. It achieves 81.1% higher total revenue, 154.8% more completed requests, and 103.6% higher average data delivery. Specifically, versus RBSM, gains reach 77.6% in revenue, 228.0% in TCR, and 242.4% in TDD; against VVMM-U: 32.6%, 43.9%, and 47.9%; versus VVMM-LW: 153.7%, 74.5%, and 284.1%; and versus VVMM-MA: 25.7%, 30.2%, and 53.7%, respectively. These improvements stem from RDA-ITU’s core mechanisms: revenue-optimized candidate sorting, dynamic V2V relay pairing, and adaptive bandwidth allocation. Prioritizing high-revenue services and facilitating efficient cooperative offloading, RDA-ITU ensures strong performance in dense mobile environments, thus promoting revenue-aware vehicular edge computing.
This work focuses on developing a compact multiband antenna to meet the growing demand for versatile and efficient radiating structures in modern wireless communication systems. A hexagonal fractal antenna is proposed and analyzed for applications such as mobile communications, WLAN, industrial, scientific and medical (ISM) bands, Wi-Fi, satellite links, radar systems, and military communications. By iteratively modifying the antenna geometry with larger hexagonal elements, the design enhances multiband behavior and improves key performance parameters including gain, S11, voltage standing wave ratio (VSWR), and radiation characteristics. The antenna is modeled using high-frequency structure simulator (HFSS)® and fabricated on a low-cost 0.8 mm FR4 substrate (εr = 4.7, loss tangent = 0.0197). Experimental validation is performed using a vector network analyzer. The antenna exhibits resonances at 1.75, 3.14, 4.23, 4.63, 5.59, 6.08, and 9.84 GHz with corresponding bandwidths of 90, 60, 60, 70, 40, 100, and 220 MHz, and achieves a peak simulated gain of 14.42 dBi. The proposed geometry enables multiband operation across several discrete frequency ranges.
This paper discusses and presents a model predictive control (MPC)-based predictive current control technique for a solar photovoltaic (PV)-integrated grid system during dynamic operation. This control technique employs extension pq (EPQ) theory to estimate reference currents and utilizes an MPC framework for tracking reference currents. Various MATLAB/Simulink simulations were conducted for solar PV generation (source disturbances) and dynamic loading. The results of the OPAL-RT OP4510 real-time simulation are also presented. A multifunctional grid-integrated converter (MFGC) integrates solar active power into the utility grid while achieving unity power factor, reactive power compensation, current balancing, and harmonic suppression. EPQ optimizes mathematical calculations, whereas MPC produces optimal gate drive signals for system converters. The EPQ-based MPCC technique offers adaptable control capabilities for dynamic operational conditions. Analytical observations present that traditional hysteresis current control results in increased switching ripples, harmonics, and losses. This result is attributed to the higher frequency of converter switch actions, which reduces converter efficiency compared to the MPC. Furthermore, the total harmonic distortion of the source current has improved (< 5%), therefore adhering to the IEEE 1547-2022 standard.
This research proposes an end-to-end procedure for arrhythmia detection based on electrocardiogram (ECG) signals using complex-valued convolutional neural network (CVCNN) incorporated with time-frequency representation. The proposed model leverages complex numbers to capture amplitude and phase information that enhances the ability of the model for detecting time-frequency variation in cardiac signals. First, signal preprocessing techniques-including normalization, wavelet denoising, and R-peak detection-are applied. Subsequently, the model extracts complex features from raw ECG data by employing the Hilbert transform to derive the analytic signal and the short-time Fourier transform (STFT) to generate a time-frequency representation. The proposed CVCNN framework effectively learns spatial-temporal features critical for detecting various types of arrhythmias. The method is evaluated using the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) Arrhythmia Database. The experimental results demonstrate that the proposed method achieves competitive performance compared with real-valued CNNs and existing approaches for arrhythmia detection under the same evaluation protocol. Specifically, the proposed method obtained an accuracy of 99.74%, precision of 99.36%, recall of 99.30%, specificity of 99.84%, and F1-score of 99.36%. A 10-fold cross-validation scheme is employed to evaluate the model's effectiveness, yielding superior performance relative to conventional methods. Overall, the proposed approach constitutes a competitive complex-valued learning framework for ECG-based arrhythmia classification under beat-level evaluation and offers a robust tool for artificial intelligence (AI)-based medical diagnostics.
High impedance faults (HIFs) present a critical challenge in power systems due to their subtle signal characteristics, which often remain undetected by conventional protection methods. These faults typically do not produce significant phase disturbances, making reliable detection difficult. However, analysis of the neutral-to-earth voltage (NEV) profile under fault conditions provides a promising alternative for fault identification. Existing approaches for detecting and classifying HIFs using NEV signals remain limited and may result in inaccurate maintenance decisions. This paper proposes a fault classification framework for multiple fault types, including HIF, three-phase fault, three-phase fault to ground, double line, double line to ground, and single line to ground, based on NEV profiles. Discrete wavelet transform is employed to extract discriminative features from the NEV signals, which are subsequently used as inputs to a convolutional neural network (CNN). For performance comparison, a long short-term memory (LSTM) network is also evaluated. To ensure robust and unbiased assessment, 5-fold cross-validation is adopted, and the models are tested under varying noise conditions of 40 dB, 30 dB, and 20 dB to examine noise resilience. Experimental results demonstrate that the CNN achieves a superior average classification accuracy of 96.93%, outperforming the LSTM, which attains 96%. The findings highlight the CNN's enhanced robustness and reliability in classifying fault types based on NEV profiles, particularly for challenging HIF scenarios under noisy operating conditions.
This paper presents an adaptive backstepping nonsingular fast terminal sliding mode controller integrated with a nonlinear disturbance observer to achieve precise trajectory tracking of robotic manipulators subject to model uncertainties and unknown time-varying disturbances. A dead-zone-based adaptive gain mechanism is introduced to dynamically adjust the control gain according to the deviation of the sliding surface, thereby enhancing robustness and reducing chattering. The proposed reaching law ensures fast, nonsingular, and adaptive convergence, suppressing high-frequency oscillations without compromising stability and the nonlinear disturbance observer enables real-time estimation and compensation of modeling errors, friction, and external disturbances for superior rejection. The semiglobal uniform ultimate boundedness of the closed-loop system is rigorously proven using Lyapunov theory. Simulation results confirm that the proposed method achieves higher tracking accuracy, faster convergence, and better robustness compared with conventional controllers.
The rapid growth of the global population has led to a substantial increase in the number of patients, while the availability of healthcare professionals has not expanded at a comparable rate. This imbalance highlights the urgent need for efficient and reliable computer-aided decision support systems that can reduce clinical workload while maintaining high diagnostic accuracy. In this study, a novel and systematically integrated artificial intelligence-based pipeline is proposed for medical image classification, combining statistical significance-driven feature ranking with evolutionary feature selection in a unified framework. The proposed pipeline consists of four sequential stages: feature extraction, ranking, selection, and classification. Features are extracted using pretrained AlexNet, ResNet-101, and GoogleNet architectures without requiring any network retraining. To ensure statistically grounded dimensionality reduction, individual features are ranked using two-sample z-test and one-way ANOVA, retaining only statistically significant features. Subsequently, nonlinear feature dependencies are captured through evolutionary optimization using binary genetic algorithm, binary differential evolution, and binary artificial bee colony (BABC) algorithms. The selected features are finally classified using k-nearest neighbors, support vector machines (SVM), and Na & iuml;ve Bayes classifiers under 4-fold cross-validation. The effectiveness of the pipeline is validated on three medical imaging datasets: chest X-ray, skin lesion, and breast ultrasound images. Results demonstrate that the AlexNet/ResNet-101-ANOVA-BABC-SVM configuration consistently outperforms alternative pipelines and baseline approaches. The proposed method achieves accuracies of 96.89%, 88.89%, and 92.71% on the respective datasets, while reducing the feature dimensionality to only 5-9% of the original feature space. Statistical analysis confirms that the performance improvements are significant compared to using only-features, features+ANOVA, features+BABC, and transfer learning-based models (p < 0.05). These results indicate that the proposed pipeline offers a statistically principled, computationally efficient, and highly generalizable alternative to end-to-end deep learning, providing a practical and robust decision support solution for medical image diagnosis across diverse imaging modalities.
The deployment of Internet of things (IoT) networks powered by renewable energy sources presents unique challenges in balancing security requirements, energy efficiency, and communication reliability. This paper presents a comprehensive multiobjective optimization framework for secure renewable energy IoT nodes that addresses fundamental trade-offs between these competing objectives. We develop a mathematical model incorporating energy harvesting dynamics, security protocols, and communication performance metrics across various environmental scenarios. The proposed framework employs a modified NSGA-II algorithm to identify Pareto-optimal configurations for different deployment contexts. Through extensive simulation analysis, we demonstrate that hybrid energy sources (solar-wind combinations) with lightweight security protocols achieve optimal balance across performance metrics. The results show that each security level increment reduces energy efficiency by approximately 7.5%, while hybrid configurations maintain 15-20% higher reliability compared to single-source alternatives across varied environmental conditions. The framework provides practical guidelines for IoT deployment strategies in sustainable applications.
Transferring knowledge from large-scale, independently pretrained image and text models to video understanding requires addressing several challenges, including maintaining generalization capabilities of models, integrating them into multimo dal architectures, and fine-tuning with temporal dynamics. This study evaluates the effectiveness of parameter-efficient fine-tuning (PEFT) techniques in transferring pretrained knowledge from two independent models for video action recognition within a simple, streamlined multimo dal fusion pipeline. Specifically, we adapt CLIP as the text branch and DINOv2 as the image branch, keeping both backbones frozen to preserve their pretrained robustness, while introducing lightweight, task-specific modules to adapt and fuse the branches with temporal dynamics. A simple fusion transformer combines the image and text branches, enabling their efficient integration with minimal training cost. We systematically evaluate the framework on widely-recognized midscale video benchmark datasets, comparing prompt-based and adapter-based PEFT techniques across different data regimes. Our results demonstrate that this combination achieves competitive performance, highlights the transferability and scalability of independent pretrained models for a targeted task, and provides practical insights for adapting large models using midscale, task specific video datasets. In particular, adaptations of the DINOv2 image encoder and CLIP text encoder improve recognition accuracy over the frozen baseline up to an average absolute gains of 3.47% across K5-KAll. Moreover, the proposed DoRA DI-NOv2 combined with an adapter-based CLIP text encoder achieves competitive state-of-the-art performance on UCF101, HMDB51, and DIVING48, consistently outperforming prior methods in few-shot scenarios and reaching up to 82.0% accuracy with K2 training examples.
The complex electromechanical structure of wind turbines, along with harsh operating conditions, poses significant challenges for precise and robust fault diagnosis. To address this challenge, an ensemble multifault diagnostic framework based on an adaptive chaotic artificial bee colony (C-ABC)-optimized support vector machine (SVM) and gradient boosting machine (GBM) is proposed. In the proposed framework, data redundancy and overfitting are reduced through a two-stage hybrid filter-transformer-based feature reduction approach using ReliefF, followed by Principal Component Analysis. The chaos function of the proposed C-ABC maintains an adaptive balance between the exploration and exploitation phases, thereby preventing premature convergence, which is a common problem in the traditional ABC, and ensuring optimal hyperparameters for the SVM and GBM classifiers. These optimized models were further stacked into an ensemble framework that leverages the individual strengths of the SVM and GBM classifiers, providing a multifault diagnostic model with a complementary decision boundary and enhanced multifault detection capability. Compared with conventional fault diagnosis methods, the proposed ensemble framework achieved superior classification accuracy (98.5%) and classification stability during experimental validation using real-world SCADA data. Pseudo cross-turbine validation using real-time SCADA data was conducted to evaluate and verify the generalization ability of the proposed ensemble framework.
Software Size Measurement (SSM) is crucial for estimating required project effort as well as budget and schedule. However, many small and medium-sized companies struggle to apply objective SSM due to limited resources and lack of expertise. This often leads to inaccurate estimates and project overruns. There is a need for practical, low-resource solutions that support these tasks without requiring expert involvement. Motivated by this challenge, this study proposes an automated software size measurement approach that formulates the measurement task as supervised regression over natural language requirements, using domain-adapted transformer models. We construct large-scale Turkish and English software engineering corpora to pretrain two models: SE-BERT and SE-BERTurk. These models are fine-tuned on a multilingual, organization-specific dataset annotated with COSMIC Function Points (CFP) and MicroM size by domain experts. We evaluate the models using various regression and classification metrics. Results show that SE-BERT improves exact match accuracy from 66.9% to 68.2% compared to BERT, while SE-BERTurk improves from 65.7% to 69.3% over BERTurk. Both models also achieve lower normalized errors than previous domain-adapted baselines (BERT_SE and RE-BERT), demonstrating superior generalization. These findings highlight the effectiveness of domain-specific pretraining for software engineering tasks and its potential to support accurate software size estimation, especially in low-resource languages like Turkish and in real-world, organization-specific contexts.
Tsunamis pose severe and time-critical risks to densely populated coastal cities, where limited warning times and infrastructure constraints demand carefully coordinated evacuation strategies. This study develops an integrated, risk-aware optimization framework that jointly considers vertical and horizontal sheltering options together with mixed pedestrian-vehicular evacuation dynamics. The proposed mixed-integer second-order cone programming (MISOCP) model simultaneously determines vertical shelter location, evacuee assignment, road-use designation for pedestrians and vehicles, and route selection under congestion, capacity, and budget constraints. Vehicle travel times incorporate congestion effects through a convex flow-dependent function, while pedestrian routing ensures convergent and conflict-free evacuation paths. A risk-minimization objective accounts for tsunami hazard levels, inundation conditions, and spatial exposure, prioritizing safer evacuation routes over purely time-based approaches. The model is applied to & Idot;stanbul's B & uuml;y & uuml;k & ccedil;ekmece district, one of the areas most vulnerable to tsunami impact following a major Marmara Sea earthquake. Using real geographic, demographic, and infrastructure data, we evaluate multiple budget and demand scenarios to examine the model's sensitivity to shelter investment decisions and varying proportions of pedestrian and vehicular evacuees. Results show that opening strategically located vertical shelters substantially reduces total evacuation risk and alleviates congestion in critical zones. Increasing vertical shelter investment provides diminishing returns beyond a moderate budget level, indicating the existence of an efficient investment threshold. Scenario analysis further reveals that mixed-mode evacuations can generate significant congestion on shared road segments; the model mitigates these effects through modal separation and speed adjustments. Overall, the proposed framework offers robust, context-aware decision support for tsunami-prone urban regions.
