Fault Diagnosis of NPC-Rectifiers Based on GWO-SVM

This paper presents a fault diagnosis method for 3L-NPC rectifier based on multi-information fusion, the source of features is divided into two parts: The accumulation of the voltage vector deviation in the SVPWM module during the half cycle of the control signal, and pulsation of d-axis current in an electrical signal after switch fault. A wind speed normalization model is proposed to process the characteristic data, taking into account the disturbance of wind speed fluctuation. Finally, the feature set is identified by the support vector machine (GWO-SVM) optimized by the Grey Wolf algorithm. Using the powerful global search capabilities of the Grey Wolf algorithm, the key parameters of the Support Vector Machine (SVM), such as the penalty factor and kernel function, and improve the classification and prediction ability of small samples. The diagnostic performance of this method and the effectiveness of the feature selection and normalized model are verified by simulation.