Scholarlink

<p>Distribution systems are constantly at risk of failure due to various factors, including lightning strikes, equipment aging, human mistakes, and breakdown of power system components. These occurrences impact the system's dependability, leading to costly repairs, lost productivity, and customer power outages. So, to reduce these issues, this work presents a peculiarity hybrid model (BOHDL) for determining and classifying distribution line faults for different topologies. The model uses fault current signal time series obtained by modeling and simulating an IEEE standard 13 bus test Network with PSCAD. It uses a convolutional neural network for extracting features from the training data automatically and a support vector machine model (having a good generalization ability). For classification, four pretrained convolutional neural network architectures are used in this work, which are AlexNet, GoogleNet, SqueezeNet, and ResNet. To enhance the classification accuracy, two parameters of SVM, i.e., Box Constraint and Kernel Scale, are optimized using the Bayesian approach. The effect of four different Kernel functions, i.e., polynomial, RBF, Gaussian, and Linear, on the SVM algorithm's performance is analyzed. The performance of the evolved model is assessed by calculating classification accuracy, kappa score, precision, sensitivity, and specificity. The simulation results obtained after training and testing validate the efficacy of the evolved model by giving 100% detection accuracy for both ring and radial topologies of power distribution systems. The proposed scheme has achieved a maximum of 99.98% and 96.89% fault classification accuracy for radial and ring topologies, respectively. This paper highlights the innovative aspects of the research, including the development of a novel hybrid model using advanced machine learning techniques for fault detection and classification, and rigorous validation of the model's performance using comprehensive evaluation metrics. These contributions position the proposed work as a significant advancement in the field of distribution system reliability and fault management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

BOHDL model: a robust framework for fault detection and classification in ring/radial distribution systems

  • 期刊论文

  • 发表日期：

    2025年1月6日

查看全文
  • Garima Tiwari,
  • Sanju Saini,
  • Minaxi

摘要

Distribution systems are constantly at risk of failure due to various factors, including lightning strikes, equipment aging, human mistakes, and breakdown of power system components. These occurrences impact the system's dependability, leading to costly repairs, lost productivity, and customer power outages. So, to reduce these issues, this work presents a peculiarity hybrid model (BOHDL) for determining and classifying distribution line faults for different topologies. The model uses fault current signal time series obtained by modeling and simulating an IEEE standard 13 bus test Network with PSCAD. It uses a convolutional neural network for extracting features from the training data automatically and a support vector machine model (having a good generalization ability). For classification, four pretrained convolutional neural network architectures are used in this work, which are AlexNet, GoogleNet, SqueezeNet, and ResNet. To enhance the classification accuracy, two parameters of SVM, i.e., Box Constraint and Kernel Scale, are optimized using the Bayesian approach. The effect of four different Kernel functions, i.e., polynomial, RBF, Gaussian, and Linear, on the SVM algorithm's performance is analyzed. The performance of the evolved model is assessed by calculating classification accuracy, kappa score, precision, sensitivity, and specificity. The simulation results obtained after training and testing validate the efficacy of the evolved model by giving 100% detection accuracy for both ring and radial topologies of power distribution systems. The proposed scheme has achieved a maximum of 99.98% and 96.89% fault classification accuracy for radial and ring topologies, respectively. This paper highlights the innovative aspects of the research, including the development of a novel hybrid model using advanced machine learning techniques for fault detection and classification, and rigorous validation of the model's performance using comprehensive evaluation metrics. These contributions position the proposed work as a significant advancement in the field of distribution system reliability and fault management.