A Semi-Supervised Learning Framework Based on EMD and SGAN for Online Stability Monitoring of DC Power Electronic Systems

The stability problems caused by converter interaction challenge the safe operation of dc power electronic systems (PESs). Online stability monitoring methods need to be developed urgently to prevent instability risk for dc PESs. The existing impedance measurement-based methods are invasive and need widely deployed measurement units, which is difficult to realize. In this article, a semisupervised data-driven online stability monitoring framework based on empirical mode decomposition (EMD) and semigenerative adversarial network (SGAN) is first proposed in dc PESs. The instability oscillation information can be extracted from complex dc-bus voltage by the EMD algorithm, and the semisupervised learning framework based on SGAN completes the training process with easily accessible unlabeled data and much less labeled data. Compared with the existing methods, the proposed method is noninvasive and greatly reduces the dependence on labeled data. In addition, the SGAN model is designed and the implementation process of the semisupervised online stability monitoring framework is elaborated. Finally, a dc PES model is built in the hardware-in-loop test platform and the effectiveness and good performance of the proposed method are verified by experimental tests.