Fault Diagnosis of Gas Insulated Switchgear Isolation Switch Based on Improved Support Vector Data Description Method

To improve the efficiency and precision of fault diagnosis for isolation switches within Gas-insulated switchgear (GIS), this study introduces an advanced technique utilizing an enhanced support vector data description (SVDD) algorithm. Initially, various operational states of the GIS isolation switch are simulated, and the corresponding vibration signals are captured. Subsequently, both the entropy and time-domain features of these signals are extracted to construct a multi-dimensional feature space. High-dimensional feature datasets are then reduced in dimensionality using the kernel principal component analysis (KPCA) method. Furthermore, the conventional SVDD algorithm is modified by incorporating a penalty factor, which allows for a more adaptable classification boundary. This adaptation not only focuses on positive samples but also considers the influence of selected negative samples on the classification hypersphere. Finally, the collected experimental data are classified and predicted. The results indicate that this GIS fault-diagnosis approach effectively overcomes the limitations of traditional methods, which are heavily dependent on training sample data and demonstrate poor algorithm generalization performance. This method is proven to be applicable for the fault diagnosis of isolation switches in GIS.