A Time-Frequency Feature Extraction Method Based on the Correlation Analysis with Its Motion Behaviors
Dong Wang, Aoyu Xie, Tingyue Tang, Nian Fang, Chenyu Zhou, Zhouruixing Wang, Leyu Qi · 2023
The amount of current signal data collected by the sensor is large. With the rapid growth of data today, how to use a small amount of effective data for fault detection is particularly important. It is of great significance to research the correlation mechanism between the time-frequency features and the motion behavior of an external fault arc during the occurrence of a DC fault arc.Firstly, the arc generation process was simulated to obtain a current signal and the characterization parameters of the image. Three features are extracted from the obtained fault arc image information: vertical arc movement on the left side, horizontal arc movement on the right side, and arc area change. By analyzing the image information of the three types of motion behaviors. The time-frequency features FI of the fault arc current signal in different frequency bands and the time-frequency features FM, FY, and FX of the images after time-frequency transform are obtained. Then the current data quantity is processed to ensure the consistent length with the image data. The obtained time-frequency features are imported into the SHapley Additive exPlanations (SHAP) eXtreme Gradient Boosting (XGboost) model to obtain the scatterplot of the feature density in different frequency bands. Finally, after comparing the Q values in different frequency bands, the corresponding frequency band with maximum Q is output.