Scholarlink

<p>The reliability and efficiency of electrical distribution systems are critical for maintaining uninterrupted power supply and minimizing disruptions caused by faults, which can lead to extensive outages and safety hazards. This paper presents a novel methodology that classifies electrical faults in distribution systems utilizing Empirical Mode Decomposition (EMD) together with machine learning techniques. An initial EMD decomposition of multivariate electrical signal data yields intrinsic mode functions (IMFs), which represent the non-stationary and nonlinear properties of the signals. Statistical properties such as mean and standard deviation are extracted from these IMFs to build a compact but informative representation of features. They are subsequently used to train a Support Vector Machine (SVM) classifier, together with other classical models as well as a one-dimensional Convolutional Neural Network (1D CNN) that are tasked with classification between fault and no-fault conditions. A thorough assessment was conducted using four different scenarios: two train-test split conditions (80:20 and 10:90) and two cross-validation schemes, namely 5-fold and 10-fold, applied across all the models. In the 80:20 ratio, the CNN recorded an accuracy, precision, recall, and F1-score of 100%, in hold-out testing as well as 5-fold cross-validation and 10-fold cross-validation. Classical classifiers such as Decision Tree, Random Forest, and K-Nearest Neighbor (KNN) achieved an accuracy of 100% in 5-fold and 10-fold cross-validation. The CNN achieved 98.89% accuracy in the 10:90 split. Random Forest and Decision Tree maintained perfect 100% classification. KNN showed the greatest sensitivity to limited training data at 88.89%, while SVM achieved 98.33% and Logistic Regression 95.56%. Overall, the proposed framework exhibited superior performance in classification, especially when little training data is available. The results justify the EMD-based framework as a robust and viable option for real-time fault detection in medium-voltage electrical distribution systems.</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

Multivariate fault classification in electrical distribution systems using empirical mode decomposition and machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月24日

查看全文
  • Mihir Sinha,
  • Mayukha Pal

摘要

The reliability and efficiency of electrical distribution systems are critical for maintaining uninterrupted power supply and minimizing disruptions caused by faults, which can lead to extensive outages and safety hazards. This paper presents a novel methodology that classifies electrical faults in distribution systems utilizing Empirical Mode Decomposition (EMD) together with machine learning techniques. An initial EMD decomposition of multivariate electrical signal data yields intrinsic mode functions (IMFs), which represent the non-stationary and nonlinear properties of the signals. Statistical properties such as mean and standard deviation are extracted from these IMFs to build a compact but informative representation of features. They are subsequently used to train a Support Vector Machine (SVM) classifier, together with other classical models as well as a one-dimensional Convolutional Neural Network (1D CNN) that are tasked with classification between fault and no-fault conditions. A thorough assessment was conducted using four different scenarios: two train-test split conditions (80:20 and 10:90) and two cross-validation schemes, namely 5-fold and 10-fold, applied across all the models. In the 80:20 ratio, the CNN recorded an accuracy, precision, recall, and F1-score of 100%, in hold-out testing as well as 5-fold cross-validation and 10-fold cross-validation. Classical classifiers such as Decision Tree, Random Forest, and K-Nearest Neighbor (KNN) achieved an accuracy of 100% in 5-fold and 10-fold cross-validation. The CNN achieved 98.89% accuracy in the 10:90 split. Random Forest and Decision Tree maintained perfect 100% classification. KNN showed the greatest sensitivity to limited training data at 88.89%, while SVM achieved 98.33% and Logistic Regression 95.56%. Overall, the proposed framework exhibited superior performance in classification, especially when little training data is available. The results justify the EMD-based framework as a robust and viable option for real-time fault detection in medium-voltage electrical distribution systems.