Research on Series Arc Fault Detection Method Based on the Combination of Load Recognition and MLP-SVM

In this study, a series arc fault detection method combining load recognition and multilayer perceptron-support vector machine (MLP-SVM) is proposed. The method addresses the issue of interfering loads on series arc fault detection and the lack of significant arc fault features in some loads. Initially, the eigenvalues of the line currents for single and mixed loads are extracted in the time domain, both during arc fault and normal conditions. Subsequently, load recognition is performed using a complex matrix calculation method. Then, a feature matrix and history matrix are created for each load. The history matrix is then used to compare the data in the feature matrix to detect any abnormalities in the eigenvalues of each in the presence of any irregularity, the line current flowing through this load will be consistently gathered throughout several cycles, and processed to obtain the eigenvalues, then fed into the MLP-SVM model for training. The classification outcomes will be achieved by means of model train. The results demonstrate that the method effectively prevents misclassification of interfering loads, resulting in improved accuracy in series arc fault detection.