Bulletin of Electrical Engineering and Informatics
Not a member yet
2885 research outputs found
Sort by
Optimal deployment of solar PV power plants as fast frequency response source for a frequency secure low inertia power grid
Full text linkModern power systems are witnessing increased uptake of solar photovoltaic power plants (SPVPPs) replacing conventional synchronous generators (SGs). SPVPPs lack any rotating parts resulting in no natural rotational inertia contribution to the grid. Reduced inertia makes the power system more dynamic, making it susceptible to frequency instability caused by minor disturbances. This problem is majorly addressed by limiting the penetration of SPVPPs to ensure a minimum level of critical inertia is maintained or by providing additional virtual inertia from an energy storage system. However, the SPVPPs can be configured to operate below maximum power point tracking (MPPT) (deloaded mode) to provide a reserve capacity that can rapidly be deployed as fast frequency response (FFR) in case of a frequency event. This paper presents a strategy to optimize the FFR capacity of a deloaded SPVPP using particle swarm optimization (PSO) algorithm. DIgSILENT PowerFactory was used to model the deloaded SPVPP and run time domain simulations. PSO algorithm was implemented using a Python script in PowerFactory. The proposed strategy was applied on a modified IEEE 39 bus test system. The results show that optimal deloading of SPVPP can help to successfully arrest frequency decline, reduce power curtailment while adhering to the prescribed constraints
Torque control of PMSM motors using reinforcement learning agent algorithm for electric vehicle application
Full text linkAs electric vehicles (EVs) demand higher performance and efficiency, precise torque control in interior permanent magnet synchronous motors (IPMSMs) becomes increasingly vital. This paper introduces a reinforcement learning (RL)-based method to optimize torque control in IPMSMs. The RL agent is trained to regulate d-axis and q-axis currents, producing stator voltages to follow the desired motor speed. The control system includes an observation vector, voltage-based actions, and a specially designed reward function. Due to the nonlinear dynamics of the motor, training the agent requires significant computational effort. MATLAB/Simulink simulations are performed to compare the RL controller with a traditional PI controller. Results indicate that the RL controller delivers quicker and more accurate performance, although additional training is necessary to minimize overshoot
Integration of deep learning algorithms for real-time vehicle accident detection from surveillance videos
Full text linkMajor road accidents have increased due to the rapid rise of vehicles on the roads due to affordability and accessibility. While minor accidents can be resolved without the need for escorting to hospitals, significant accidents that involve the deployment of airbags necessitate the immediate attention of authorities. Thus, subsequent action of first aid and proper communication to concerned medical personnel can avoid most fatalities from accidents. The system involves the automatic detection of traffic accidents from videos extracted by closed-circuit television (CCTV) surveillance. In case of an accident, the system will detect and information about the accident will be instantly relayed to the nearest medical center. We have implemented different machine learning models such as Resnet-18, VGG-16, LeNet, and Inception V1 to ensure the accuracy of accident detection. From comparing all these models, the convolutional neural network (CNN) model shows the highest accuracy of 98%. The quick response will be an important step toward a safer and more secure transportation landscape
Detecting spam using Harris Hawks optimizer as a feature selection algorithm
Full text linkThe Harris Hawks optimization (HHO) was used in this study to enhance spam identification. Only the features with a high influence on spam detection have been selected using the HHO metaheuristic technique. The HHO technique's assessment of the selected features was conducted using the ISCX-URL2016 dataset. The ISCX-URL2016 dataset has 72 features, but the HHO technique reduces that to just 10 features. Extra tree (ET), extreme gradient boosting (XGBoost), and support vector machine (SVM) techniques are used to complete the classification assignment. 99.81% accuracy is attained by the ET, 99.60% by XGBoost, and 98.74% by SVM. As we can see, with the ET, XGBoost, and k-nearest neighbor (KNN) techniques, the HHO technique achieves accuracy above 98%. Nonetheless, the ET technique outperforms the XGBoost and KNN techniques. ET outperforms other methods due to its robust ensemble approach, which benefits from the diverse and relevant feature subset selected by HHO. HHO's effective reduction of noisy or redundant features enhances ET's ability to generalize and avoid overfitting, making it a highly efficient combination for spam detection. Thus, it looks promising to combat spam emails by combining the ET technique for classification with the HHO technique for feature selection
Arecanut grading classification based on representational deep neural network with support vector machine
Full text linkThe grading of arecanuts before their sale is significant for enhancing profitability. The assessment of areca nut quality widely utilizes and respects both producer-level and wholesale dealer-level grading methods. This study proposes an advanced grading framework for white Chali-type arecanuts by developing a standardized image database and utilizing deep learning-based feature extraction. This research presents a novel approach by combining a representational deep neural network (ResNet) for automatic feature extraction with various spectral analysis methods, such as the Fourier transform and wavelet transform, to capture frequency-domain features. The support vector machine (SVM) model classifies these extracted features. The proposed system achieves an accuracy of 97.8%, which is significantly better than existing methods SVM with 72.5%, convolutional neural network (CNN) with 92.9%, AlexNet with 90.6%, and VGG19 with 90.2%. The results show that the proposed hybrid ResNet-SVM method improves accuracy, precision, recall, and F1-score, making it a more reliable and automated way to grade areca nuts. This method thus enhances efficiency, reduces manual effort, and ensures consistent quality assessment
Enhanced autonomous water garbage collection system using deep learning-based object detection and path planning
Full text linkWater pollution, particularly from floating debris such as plastics, has become a critical environmental issue, threatening aquatic ecosystems and biodiversity. Autonomous solutions for the detection and removal of waste are increasingly essential for maintaining water cleanliness and mitigating pollution. However, existing systems face limitations in real-time detection, accuracy, and adaptability to diverse aquatic environments. This paper utilizes the water pollution images dataset, comprising almost 300 high-resolution images from lakes, rivers, and coastal areas, representing various types of floating waste under different environmental conditions. In response to these challenges, this paper introduces an autonomous unmanned surface vehicle (USV) system equipped with the enhanced waste detection network (EWD-Net). EWD-Net improves upon traditional single-shot detection algorithms by integrating deeper feature extraction layers and enhancing computational efficiency, resulting in higher accuracy and faster detection. Additionally, the system includes the dynamic path optimization (DPO) module for efficient navigation and obstacle avoidance in complex water environments. The novelty of this system lies in its dual approach, combining advanced detection with optimized path planning, ensuring effective autonomous operation. The results indicate that the proposed model achieves an accuracy of 94.6%, outperforming existing algorithms and providing a robust solution for real-time waste detection and collection
A comparative study of machine learning methods for drug type classification
Full text linkDrugs, commonly called narcotics, are dangerous substances that, if consumed excessively, can result in addiction and even death. Drug abuse in Indonesia has reached a concerning stage. In 2017, the National Narcotics Agency detected 46,537 drug-related incidents, including methamphetamine, marijuana, and ecstasy. There are 4 types of substances that can affect drug users, such as hallucinogens, depressants, opioids, and stimulants. A machine learning approach can detect these substances using user symptom data as input. This study uses six different methods in classifying, including decision tree, C.45, K-nearest neighbor (KNN), random forest, and support vector machine (SVM). The dataset comprises 144 data and 21 attributes based on the user's symptoms. The evaluation method in this study uses cross-validation with K-fold values of 5 and 10 and uses three parameters: precision, recall, and accuracy. KNN yields the most optimal results by using K=1 and K-fold 10 in the Euclidean and Minkowski types. The model achieves precision, recall, and accuracy of 91.9%, 91.7%, and 91.67%, respectively
Ensemble and deep learning via median method for learning disability classification
Full text linkThe study explores the classification of students with and without learning disabilities (LD) through machine learning techniques, utilizing a real dataset and implementing bootstrapping for data augmentation. Noteworthy findings reveal the Adam optimizer's superior performance among various optimizers, achieving a true positive rate (TPR) of 0.97 and a false positive rate (FPR) of 0.02, with high precision, recall, and f1-score values. Additionally, ensemble learning, employing the median method, combines models like Random-ForestClassifier and KerasClassifier, and BaggingClassifier with KerasClassifier, resulting in improved performance. However, the Median-Combined model, integrating AdaBoostClassifier and KerasClassifier, stands out with an accuracy of 99.6%, along with elevated precision, recall, and f1-score values. The comprehensive classification report showcases an overall FPR of 0.0 and TPR of 0.999, highlighting the enhanced performance of the combined model. The significance of this study lies in underscoring the power of fusion between ensemble learning and deep learning techniques, leveraging the median method. This combined model exhibits superior performance, excelling in accuracy, precision, recall, and overall classification effectiveness. The innovative approach of combining both ensemble and deep learning methods through the median method not only advances the understanding of learning disability classification but also emphasizes the practical importance of integrating diverse methodologies for enhanced model performance
HBRFE: an enhanced recursive feature elimination model for big data classification
Full text linkThe process of classification in big data is a tedious task due to the large number of volumes, veracity, and variety of the data. Classification of big data pave the path to organize the data and improve the classifier performance. This research article proposed a Hadoop framework based recursive feature elimination-based model called HBFRE for extract significant features from the big data by integrating map and reduce frame work. HBFRE extract the significant features by removing the least and irrelevant features from the dataset by using refined recursive feature elimination (RFE) with map and reduce framework. This method takes the mean of each attribute and find the variance in each instance. The proposed model is evaluated and analyzed by the accuracy performance and time complexity. This research utilized various classifier like artificial neural network (ANN), support vector machine (SVM), random forest (RF), k-nearest neighbors (KNN), and AdaBoost to measure the classification performance on the big data. Proposed HBRFE model is compared with different feature selection like RFE, relief, backwards feature elimination, maximum relevance k-nearest neighbors (MR-KNN), and scalable deep ensemble framework big data classification (SDELF-BDC)
Enhancing low-light pedestrian detection: convolutional neural network and YOLOv8 integration with automated dataset
Full text linkThis research aims to enhance the you only look once (YOLO) model for pedestrian detection in environments with varying lighting conditions, particularly in low-light scenarios. The primary contribution of this work is the integration of a convolutional neural network (CNN)-based low-light enhancement model, which transforms dark images into brighter, more discernible ones. This enhanced dataset is subsequently used to train the YOLO model, allowing it to learn from both the original and transformed data distributions. Unlike traditional YOLO training approaches, this method generates more accurate data representations in challenging lighting environments, leading to improved detection outcomes. The novelty of this approach lies in its dual-stage training process, which integrates a CNNbased low-light enhancement model with YOLO’s detection capabilities. This combination not only enhances pedestrian detection but also has the potential for application in other domains, such as vehicle detection and surveillance, particularly in challenging lighting conditions. The automatic dataset collection pipeline provides an efficient way to gather diverse training data across various scenarios. The YOLOv8 model trained on the low-light enhanced dataset significantly outperformed the baseline model trained only on the original dataset, with precision increased by 9.8%, recall by 45.7%, mAP50 by 26.8%, and mAP50-95 by 41.0% when validated on dark images