부산대학교 도서관

It is difficult to detect the residual current of specifi c fault types in low-voltage distribution networks, which results in few labeled residual current samples. Thus, it is difficult to recognize the fault types of residual current. To solve this problem, a cooperative training classification model based on an improved squirrel search algorithm (ISSA) for a semi-supervised support vector machine (S3VM) and the k-nearest neighbor (KNN) is proposed (ISSA-S3VM–KNN). First, the residual current is decomposed into k intrinsic mode functions (IMFs) by variational mode decomposition (VMD), and the characteristic parameters of the IMFs are extracted to obtain a characteristic dataset for establishing a classifi cation model. Second, to solve the problem where it is difficult to the select parameters (such as the penalty factors, slack variables and kernel function) of a S3VM, an ISSA parameter optimization method is proposed to self-adaptively select the optimal combination of parameters for the S3VM. Finally, the KNN is used to verify the classification results of an ISSA-S3VM through cooperative training, which further improves the classification accuracy of the S3VM for unlabeled residual current samples. Classifi cation results of measured and simulation data show that the classification accuracy of the ISSA-S3VM–KNN is higher than that of the SVM–BPNN, WE–AE–BPNN, and PSO–SVM. The ISSA-S3VM–KNN provides a certain theoretical basis for achieving fast and accurate residual current fault type recognition.