Arc Fault Detection of Electric Bicycle Based on FTS- DPC

Arc fault is one of the important causes of electric bicycle fires. However, the current research on arc fault detection for electric bicycles is not mature enough. In this paper, we propose an arc fault detection method based on multi-domain arc features and improved Density Peak Clustering. Firstly, the current signal of arc fault in the charging state of electric bicycle is collected, and the Tamura texture feature is extracted after the signal is two-dimensional imaged by Markov Transition Field method. In order to avoid the instability of a single feature, time domain and frequency domain features are extracted as supplements. Secondly, due to the randomness of arc, the feature value of arc samples fluctuates greatly and overlaps with the normal samples, which will decrease the detection accuracy. Therefore, an improved Density Peak Clustering method based on Fuzzy Threshold Screening is proposed. The optimal fuzzy threshold is obtained by training, according to the threshold screening results, the feature values of normal samples and arc samples are mapped to 0 and 1 respectively, which increases the inter-class difference and reduces the intra-class difference. Finally, the arc fault detection is realized based on the Density Peak Clustering method. The test accuracy in the experimental results is 99.4%, which is higher than other arc fault detection methods.