Identifying Distribution Network Faults Using Adaptive Transition Probability

TL;DR

This paper introduces an adaptive transition probability method that enhances fault detection accuracy in distribution networks by aligning simulated and real data features, outperforming traditional classifiers especially with limited samples.

Contribution

It proposes a novel adaptive probability learning approach combined with waveform decomposition to improve fault detection accuracy in distribution networks.

Findings

Outperforms CNN, SVM, and KNN in fault classification accuracy.

Effectively handles limited sample sizes in real-world data.

Demonstrates robustness under adaptive learning conditions.

Abstract

A novel approach is suggested for improving the accuracy of fault detection in distribution networks. This technique combines adaptive probability learning and waveform decomposition to optimize the similarity of features. Its objective is to discover the most appropriate linear mapping between simulated and real data to minimize distribution differences. By aligning the data in the same feature space, the proposed method effectively overcomes the challenge posed by limited sample size when identifying faults and classifying real data in distribution networks. Experimental results utilizing simulated system data and real field data demonstrate that this approach outperforms commonly used classification models such as convolutional neural networks, support vector machines, and k-nearest neighbors, especially under adaptive learning conditions. Consequently, this research provides a fresh perspective on fault detection in distribution networks, particularly when adaptive learning conditions are employed.