This work presents a hybrid Taguchi-based dung beetle optimization (TDBO) method to resolve the optimal power flow (OPF) issue in a transmission network. The TDBO method hybridizes the heuristic search capabilities of the DBO with the exploration capabilities of the Taguchi method (TM). This hybridization enhances the solution accuracy, convergence rate, and initial population diversity. The DBO mimics dung beetle breeding, ball-rolling, dancing, foraging, and thievery. The OPF study is formulated with four objective functions, including total generation cost, emission, active power loss, and voltage magnitude deviation, while fulfilling equality and inequality constraints. To address several constraints in the OPF issue, the superiority of a feasible solution (SF) - based constraint-handling approach is deployed. Additionally, a novel TOPSIS-based decision-making strategy is employed to obtain the best-compromised solution (BCS) from the Pareto optimal (PO) front. The TDBO-SF method is validated on the IEEE 30-, 57-, and 118-bus test systems, and its performance is compared with other methods. The TDBO-SF method attains a generation cost of 841.7231/h and an emission of 0.2366ton/h in Case 5, a generation cost of 42179.80/h and an emission of 1.2615ton/h in Case 12, and a generation cost of 135716.72/h and an active power loss of 34.5983 MW in Case 15. The outcomes corroborate that TDBO-SF improves convergence rate, accuracy, and computing time.

Photovoltaic (PV) systems have gained importance as one of the major renewable energy technologies because of its clean and sustainable characteristics; however, PV nonlinear current-voltage and power-voltage characteristics are a significant factor in limiting the maximum available power determined by the amount of available power under different environmental and load conditions. Maximum Power Point Tracking (MPPT) algorithms are one of the most important performance enhancements to help make PV energy conversion systems more efficient and reliable. Conventional MPPT techniques such as Perturb and Observe (P&O) have been widely used due to their simple and easy implementation in terms of control strategies, however, they exhibit steady-state oscillations, slow convergence, and poor dynamic performance under fast changing conditions of irradiation and load. To overcome these limitations, validation of intelligent and bio-inspired optimization and MPPT algorithms have attracted a great deal of interest. This paper includes a comprehensive comparative analysis of three MPPT techniques that are: conventional P&O algorithm, Grey Wolf Optimization algorithm (GWO), and Teaching-Learning-Based Optimization algorithm (TLBO). A detailed model of PV system coupled with dc-dc boost converter is built in Matlab/Simulink and the algorithms are tested with a constant irradiation and variable load and with simultaneous irradiation and load variation. Performance criteria like tracking performance, speed of convergence, steady state oscillations and robustness under dynamic conditions are analyzed. Based on the simulation results it is proven that both GWO and TLBO clearly outperforms the conventional P&O algorithm. Amongst all the optimization-based approaches, TLBO provides good performance with near-instantaneous convergence, minimum oscillations and consistent tracking efficiency is close to 100% for all the tested scenarios. The results prove that TLBO-based MPPT offers solid and computationally efficient solution for real world applications of PV systems especially in those with frequent and random operating condition variation environment.

Microgrid integration introduces fast, stochastic disturbances that challenge frequency stability. This paper presents a two-degree-of-freedom fractional-order proportional tilt integral derivative plus one controller (2DOF-FOPTID+1) tuned with a Modified Walrus Optimization Algorithm (MWA) to mitigate frequency deviations while preserving tracking performance. The novelty lies in jointly deploying a 2DOF-FOPTID+1 structure for decoupled tracking and regulation, an MWA-based tuning strategy tailored for resilient frequency control, and the explicit use of aggregated electric vehicles as fast distributed storage to damp frequency and tie-line power excursions; hardware-in-the-loop validation using an OPAL-RT platform is included to demonstrate practical feasibility. The controller is evaluated under step and random load variations, and robustness is examined for ±25% parameter perturbations and stochastic renewable inputs. Compared with the strong baselines PID, FOPID, 2DOF-PID, and FOPTID, the proposed approach reduces settling time by up to 39.27% and lowers peak-to-peak frequency deviation by about 20.88% under these operating scenarios, indicating a practical and effective solution for enhancing frequency resilience in microgrid-integrated power systems.

The utilisation of wind energy has attracted considerable interest due to its potential as a sustainable and ecologically sound form of power generation. Nevertheless, the accurate prediction of wind speed continues to pose a significant challenge, primarily due to the inherent variability and randomness associated with this natural phenomenon. Transformer-based models have been increasingly employed for wind speed prediction (WSP), yet they encounter challenges in handling past input data redundancies and optimizing encoder-decoder connections. To tackle these issues, a novel hybrid model, SVMD-YFormer-KMSE is presented. The proposed model integrates successive variational mode decomposition (SVMD) with a Y-former model, using the K-MSE loss function. The use of the SVMD component yields enhanced data decomposition and concurrently enhances computing efficiency. SVMD is employed to denoise the input data and then denoised data inputted into the Y-former model to interpret both coarse and fine-grained characteristics for accurate WSP. The Y-former model integrates sparse attention modules to interpret the wind speed efficiently. Moreover, the Y-former model also maintains stability within encoder and decoder stacks through the reconstruction of recent past data, thereby ensuring consistency and reliability in WSP forecasts. The proposed Kernel-MSE loss function effectively captures the nonlinear nature of wind speed data, enhancing model accuracy and convergence. The proposed hybrid model has been extensively evaluated using Leicester and Portland wind farm data, showcasing its resilience and capacity to generalise over different wind speed profiles.

As the global population grows, so does the demand for power. When demand exceeds generation, system frequency decreases, leading to instability. Both single-area and multi-area power systems face frequency stability issues. Load frequency regulation is key to maintaining reliable power delivery. In this work, a fractional order proportional tilt integral derivative (FOPTID) is developed for the load frequency control issue for a two-area interconnected multi-source power system. The efficacy of the FOPTID controller is compared with the traditional controllers like PID and FOPID by adapting it to the two area system. The designed system is then tuned using a recent optimization, i.e., sea horse optimization (SHO), for optimally deriving the optimal controller parameters. The vehicle model (EV) is incorporated in area-2 so that the frequency profile can be enhanced. Finally, it is observed that the SHO-tuned FOPTID controller performs better and produces optimal responses. The proposed FOPTID with the electric vehicle model results in minor transients in frequency and tie-line responses and settling of frequency to the desired 50Hz at a quicker time.

Accurate wind speed prediction is crucial for optimizing wind energy generation and ensuring effective integration into power systems. Reliable forecasts enable better adaptation of power grids to renewable energy sources, refining overall efficiency and sustainability. This study explores the use of machine learning techniques for wind speed forecasting by using meteorological data, such as temperature, air pressure, wind direction, and height. Multiple regression models, which would include K-Nearest Neighbours (KNN), Random Forest, Decision Tree, Support Vector Regression (SVR), and Linear Regression are compared to assess the models Mean Squared Error (MSE) and R2 scores of a real-world wind speed datasets to predict wind speed. Among all Random Forest model performed the best, proving its adaptability in identifying complex relationships with MSE values of 0.27 and 0.998 respectively, KNN and Decision Tree also formed modest results. The results demonstrate how ensemble-based models, such as Random Forest, can accurately forecast wind speed, which makes them appropriate for optimizing renewable energy. In order to improve wind speed prediction for energy applications, this study offers insights into model performance and potential paths for future research.

In this work, a novel bio-inspired multi-objective artificial hummingbird algorithm (MOAHA) is proposed to address the optimal power flow (OPF) issue in a renewable energy and plug-in electric vehicle (PEV) integrated power system. The MOAHA mimics the unique flight abilities and intelligent foraging approaches of hummingbirds in their natural habitat. The proposed method is assessed by resolving the OPF with multiple objectives using the IEEE 57-and IEEE 118-bus systems. The objectives include minimizing the total generation cost, emission, active power loss, and voltage magnitude deviation along with various constraints. This work employs a superiority of the feasible solution (SF) based constraint handling methodology to address a variety of constraints associated with the OPF study. Additionally, a TOPSIS approach is deployed to identify the best trade-off solutions among multiple objectives in the multi-objective OPF (MOOPF) problem. These techniques further aid in optimal outcomes to be specific the proposed MOAHA-SF attains a generation cost of 35448.14$/h and emission of 0.8696ton/h in the modified IEEE 57-bus system, and 127890.64$/h for cost and 31.4611 MW for active power loss in the modified IEEE 118-bus system. As demonstrated by the results, the proposed MOAHA-SF is significantly more effective in terms of computation efficacy and solution accuracy. Furthermore, the performance of the MOAHA-SF is evaluated through the calculation of hypervolume metrics and a one-sided ANOVA.

Handwritten character recognition (HCR) is still regarded as a difficult learning problem in pattern recognition, even after being studied in-depth for a few decades. Research on script independent models is also scarce. This can be linked to several things, especially character structure similarities, variances in handwriting styles, noisy datasets, script diversity, the conventional research's emphasis on manual feature extraction techniques, and the lack of publicly available datasets and code repositories to replicate the findings. However, deep learning offers from top to bottom learning together with has achieved great success in various pattern recognition domains, such as hand gesture recognition (HCR). Deep learning methods, on the other hand, are computationally costly, require a lot of data to train, and are limited to scripts. We have developed a novel generic Deep Progressively Learning Model architecture for script independent handwritten character recognition, known as HCR-DPLM, to overcome the limitations. The foundation of HCR-DPLM is a unique transfer learning strategy for HCR that makes use of a pre-trained network's feature extraction layers in part. HCR DPLM offers better performance and better generalizations, faster and computationally efficient training, and the ability to work with small datasets because of Transferring Knowledge and Imagery augmentation. In this paper, we use a pixel level classifier to extract the character pixels and eliminate noise from handwritten character images.

Modern energy storage systems, especially those in electric vehicles, rely heavily on lithium-ion batteries. These power sources are favoured for their high energy density, extended lifespan, and adaptability to various applications. Accurate estimation of their State of Health (SoH), a key metric for assessing capacity degradation over time, is critical for ensuring safety, performance, and efficient battery management. The Battery Management System (BMS) mitigates risks such as temperature variations, overcharging, and over-discharging by continuously monitoring metrics like State of Charge (SoC) and SoH, ensuring optimal battery utilization and reliability. This study evaluates least square methods—WLSM, PLSM, FMPLSM, and geometric method for online SoH estimation, demonstrating geometric method’s superior accuracy (error < 2%) on real-time EV battery data. Special emphasis is placed on online SoH estimation, enabling real-time battery monitoring and management during vehicle operation. These advancements maximize the utility of energy storage systems and support the transition toward sustainable, eco-friendly mobility solutions. By enhancing SoH estimation accuracy, the Geometric Least square Method contributes to improved battery performance, extended operational lifetimes, and better energy management systems.

The Internet of Vehicles (IoV) refers to a network where vehicles are connected to each other and to the surrounding infrastructure, enabling the exchange of information among vehicles and with the infrastructure to improve safety, efficiency, and overall transportation experience. However, intrusion into IoV systems may compromise the privacy of vehicle owners and occupants, with unauthorized access to personal data collected by vehicles leading to privacy violations and identity theft. Anomaly-based Network Intrusion Detection System (A-NIDS) provides a means to recognize prospective threats by detecting deviations from the normal model. This work introduces an intrusion detection system for vehicles using Autoencoder and Decoder for Image-Based detection, offering automated operation without supervision.

This research study explores the application of pre-trained Convolutional Neural Network (CNN) architectures for the classification of lung and colon cancer using histopathological images. Leveraging the LC25000 dataset, which contains 25,000 high-quality images across five categories, the research employs EfficientNetB6, ResNet34, VGG-19, MobileNetV2, and ResNet50 for image classification. Preprocessing techniques, feature extraction, and classification were carried out using CNN-based transfer learning, optimizing computational efficiency and accuracy. The results reveal that MobileNetV2 and ResNet50 outperform other models with an accuracy of 99%, followed by VGG-19 at 98% and ResNet34 at 97%. EfficientNetB6 achieved 93% accuracy, showcasing reasonable performance despite being less computationally intensive. Evaluation metrics, including precision, recall, F1 score, and accuracy, confirmed the models' efficacy in distinguishing cancerous and normal tissues. The findings demonstrate the ability of deep learning as a robust and efficient tool for early and accurate detection of lung and colon cancer, offering significant support for clinical decision-making.

Lung cancer is a significant global health issue, and early detection of lung nodules is vital for timely diagnosis and treatment. Computed Tomography (CT) scans are instrumental in classifying malignant and benign nodules. This study explores the application of ensemble learning algorithms, specifically AdaBoost, XGBoost, CatBoost, and LightGBM, to increase the accuracy of lung malignancy identification. By assessing these models using criteria such as accuracy, sensitivity, and specificity, and specificity, the analysis demonstrates that both LightGBM and CatBoost individually achieve high performance, with accuracies of 97.11% and 97.88%, respectively. LightGBM shows a sensitivity of 96.3% and specificity of 97.64%, while CatBoost slightly outperforms it with a sensitivity of 97.28% and specificity of 98.28%. The combined use of LightGBM and CatBoost yields the highest efficacy of 99.34% accuracy, 99.52% sensitivity and 99.13% specificity. This ensemble approach leverages the strengths of both models, significantly enhancing detection accuracy and reliability, thereby presenting a highly effective solution for lung cancer detection.

The nine-level cascaded multilevel inverter is presented in this article. It is mainly used for medium voltage and medium power applications. The main intention of this research work is to generate nine level ac output with a dc power excitation. In this topology, the number of power switches is reduced. The phase deposition pulse width modulation technique is implemented in this topology for nine level ac voltage generations. The proposed inverter was tested and results were validated. The harmonic analyses of the proposed inverter topology with three control techniques are illustrated. The total harmonic distortions of the inverter with various control techniques are compared..

Integrating Deep Learning-based Predictive Maintenance (DL-PM) in Industrial Internet of Things (IoT) applications optimizes operational efficiency and reduces downtime. This reaserch work explores the synergy between deep learning and predictive maintenance within the Industrial IoT context. The study traces the evolution from traditional maintenance to modern deep learning methods facilitated by IoT's real-time data acquisition. Methodologically, it details data preprocessing, model selection, and design for DL-PM. Convolutional Neural Networks (CNNs) analyze sensor data, Recurrent Neural Networks (RNNs) predict time-series patterns and hybrid models incorporate transfer learning.The reaserch work demonstrates DL-PM's application across industries through diverse case studies, evaluating its performance and comparing it to conventional approaches. It highlights challenges like data quality, model interpretability, scalability, ethical concerns, and biases. Future directions encompass advanced deep learning techniques, edge-cloud integration, collaborative learning, and strategies to overcome challenges. DL-PM's potential to revolutionize industrial processes in the IoT era is emphasized.This reaserch work underscores the transformative impact of Deep Learning-based Predictive Maintenance in Industrial IoT. It provides insights into implementation, challenges, and prospects, guiding industries towards efficient, downtime-minimized operations through DL-PM integration.

Chronic kidney disease (CKD) establishes substantial health risks, potentially progressing to life-threatening stages necessitating dialysis or surgery for survival. Early detection and effective management are crucial in mitigating its progression. This study employs ensemble learning algorithms-AdaBoost, XGBoost, CatBoost, and LightGBM-to enhance diagnostic accuracy and refine patient management strategies for predicting CKD. Utilizing the CKD dataset from the UCI Machine Learning Repository, the models are evaluated through crossvalidation, focusing on metrics such as accuracy, sensitivity, and specificity. The combined AdaBoost & CatBoost model emerges as highly effective, achieving precision of 0.99, accuracy of 99.9%, and sensitivity of 1.0. This highlights a synergistic synergy between AdaBoost and CatBoost, underscoring their potential in enhancing predictive capabilities for CKD, thereby offering promising avenues for improving clinical outcomes through early intervention and targeted therapy.

Intrusion Detection Systems (IDS) plays a major role in modern network security strategies, offering various methods and architectures to analyze network access. These systems can be roughly labelled as two categories: signature-based and anomaly-based. Signature - based IDS monitor events using a database of known intrusions, while passive IDS focus on understanding system behavior and identifying anomalies. However, with the rapid development of the IoT, new and complex security challenges possess emerged. Despite efforts to address IoT cybersecurity through various technologies, further development is essential to effectively safeguard IoT ecosystems. A well-known approach to enhance IoT security involves integrating machine learning techniques. Numerous studies have explored the application of deep learning and machine learning methods to improve Internet of Things security. This research study has developed a deep learning based method to detect attacks on IoT systems. By employing Python programming and tools such as Tensorflow, Scikit-learn, and Seaborn, the efficiency of deep learning models is utilized in enhancing detection accuracy. The resultant findings suggest that deep learning holds significant promise for enhancing IoT security measures, providing a more robust defense against cyber threats targeting IoT devices and networks. This research study has contributed to enhancing the arena of IoT security, addressing a critical need in the constantly changing field of cybersecurity.

Acute lymphoblastic leukemia (ALL) is a form of leukemia depicted by the rapid proliferation of adolescent white blood cells (WBCs) in the bone marrow. The incidence rate of ALL is approximately 80% in children and 40% in adults. This condition disrupts the production of normal cells, causing neurological abnormalities and potentially leading to fatal outcomes. Consequently, prompt and precise diagnosis is crucial for efficient therapy and recovered survival rates. This study investigates the efficacy of various machine learning models, including XGBoost, Adaboost, LightGBM, and CatBoost, in the prediction of Acute Lymphoblastic Leukemia (ALL). Through comprehensive data preprocessing, feature selection, and model tuning, we assess the performance of these models' using accuracy, precision, recall, and F1 score metrics. Our findings reveal that XGBoost significantly outperforms the other models, achieving the highest accuracy (99.31 %), precision (95.63 %), recall (98.01 %), and F1 score (99.35%). LightGBM also demonstrates strong performance, particularly in precision and recall, making it a viable alternative. Although Adaboost and CatBoost provide satisfactory results, they lag behind XGBoost and LightGBM. This study underscores the importance of selecting advanced machine learning algorithms to enhance diagnostic accuracy for ALL, ultimately contributing to improved patient outcomes through more precise and reliable detection methods.

In the modern distribution management system, state estimation (SE) has a key role in monitoring the entire distribution system using advanced measurement units. Since the conventional weighted least squares (WLS) approach faces convergence issues due to insufficient measurement devices in the distribution system, deep learning (DL) models are highly effective in performing SE. However, the existing DL models suffer from generalizability issues, unawareness of the system’s physics, and scalability issues. To address this, physics-based deep learning models have been introduced. In this paper, a physics-based temporal convolutional neural network (Ph-TCN) is proposed to perform distribution system SE (DSSE). The first stage of the proposed approach utilizes a conventional TCN model in a supervised manner for the estimation of system states, while the latter stage incorporates physics-based properties to reconstruct the measurements. The use of the TCN model in first stage facilitates the capturing of temporal features with a large receptive field, whereas the inclusion of physics equations in the second stage with problem-specific Huber-loss function enhances the performance of the model to provide accurate SE results. Simulation studies are performed on modified IEEE 13-node and IEEE 37-node distribution test systems and the corresponding results have been compared with the WLS approach and existing DL models. The numerical results show that the proposed Ph-TCN approach has exhibited superior performance and its robustness is also tested under the presence of different percentages of missing measurement data. Finally, the models’ performance is also evaluated for the modified IEEE 123-node distribution system to check for the scalability issues.

This research investigates the application of machine learning techniques, specifically neural networks, to estimate the critical voltage of flashover in external insulators. The proposed method utilizes experimental data and theoretical models to train and validate the neural network. By analyzing factors such as environmental conditions, insulator design, and operating parameters, the model can predict the critical flashover voltage with high accuracy. The integration of IoT technology enables real-time monitoring of insulator conditions, facilitating timely maintenance and reducing the risk of power outages. This research contributes to the advancement of power system reliability and enhances the safety and efficiency of electrical grids.

Diabetic foot ulcers (DFUs) are prolonged wounds that commonly affect individuals with diabetes, often resulting from a combination of neuropathy and poor circulation. These ulcers pose significant health risks and can lead to severe complications, including infection and amputation, if not properly managed. This paper presents an advanced approach to diabetic foot ulcer (DFU) classification and recognition using transfer learning models. This study evaluates the performance of various transfer learning models in the classification of Diabetic Foot Ulcers (DFUs), focusing on their accuracy, specificity, and sensitivity. The models assessed include ResNet152, EfficientNetB7, and SE-ResNeXt. EfficientNetB7 achieved the highest accuracy at 99.65% and sensitivity at 99.8%, indicating its superior capability in accurately identifying DFU cases and true positives. ResNet152, with an accuracy of 99.39%, demonstrated the highest specificity at 99.8%, effectively minimizing false positives. SE-ResNeXt, while slightly lower in performance metrics, still provided robust classification capabilities. These results highlight the efficacy of transfer learning models in medical image classification, emphasizing their potential to improve diagnostic accuracy and patient outcomes in DFU detection.