부산대학교 도서관

A series arc fault can easily ignite the surrounding flammable objects, leading to safety hazards. Therefore, accurately detecting the arc fault is essential. However, the fault characteristic information carried by the current when series arc faults occur is easily masked by the so-called screening load. As a result, series arc faults are not easily detected. To address this problem, this article proposes an arc fault detection model based on ${L}2/{L}1$ norm and classification algorithm. ${L}2/{L}1$ norm is introduced to quantify the fluctuations in the current signal when an arc fault occurs, and then combined with some commonly used time domain indexes and frequency domain indexes to extract the features of the arc fault current. Next, the extracted features are filtered to create a high-quality dataset. Subsequently, a random forest model is constructed and the dataset is used to train and test the model. Finally, the parameters of the random forest model are optimized using the grid search method to obtain a highly accurate arc fault detection model. The effectiveness of the introduced feature is verified using the arc fault data under different loads. Meanwhile, the proposed method is tested and compared with seven commonly used machine learning methods, which reflects the superiority and accuracy of the proposed method.