Open-Circuit Fault Diagnosis for a Modular Multilevel Converter Based on Hybrid Machine Learning
Yang An, Xiangdong Sun, Biying Ren, Xiaobin Zhang
Abstract
With the wide application of a modular multilevel converter in various power conversion fields, submodule open-circuit fault diagnostics have attracted increasing attention, as some of the existing diagnosis methods have a single function and limited localization speed. Therefore, a simplified and innovative multifunctional hybrid machine learning-based fault diagnosis strategy for the submodules is proposed. Starting from the output characteristics of the faulty submodule, the eigenvalues of the bridge arm current and submodule capacitor voltage during faults are extracted, and the eigenvalues are utilized for fault detection and location via the integration of improved supervised learning and unsupervised learning. Finally, the effectiveness of the proposed method is verified by simulated and experimental results in a three-phase modular multilevel converter topology. In addition, it can diagnose multiple fault types and achieve a high fault identification probability.