A Robust Unsupervised Stability Monitoring Method Based on CACSN to Address Oscillation Variations in Harmonics-Rich DC Power Electronic Systems

Stability monitoring methods are important for dc power electronic systems (PESs) to reduce instability threats that caused by interactions between converters. The initial stability monitoring method is impedance-ratio-based evaluation method, which needs to invade the system to measure the converter’s impedance and affect system’s original operation. As a result, the data-driven-based noninvasive stability monitoring method was proposed to extract instability features from dc bus voltage without impedance measurement. However, if the dc PESs suffer rich background harmonics in bus voltage or its instability oscillation frequency is changed, the existing data-driven based stability monitoring method may fail to work due to lack of labeled data. To address these issues, a robust unsupervised online stability monitoring method based on convolutional-attention-contrastive-siamese-network (CACSN) is proposed in this article. The proposed method utilizes a hierarchical temporal static CNN and hierarchical temporal dynamic attention mechanism (AM) to extract features from present, future and past instability information, ensuring the performance under varying instability oscillation and background harmonics. Additionally, stability-monitoring-target-driven contrastive learning with interpretable label mapping is introduced to enable label-free training. Finally, the effectiveness of the proposed method is validated by 3 different dc PES experimental testing platforms.