Research on transformer voice print pattern anomaly diagnosis algorithm based on gaussian mixed model and blind source separation

To extract the voice print signal of the transformer winding and iron core, and to achieve anomaly diagnosis of transformer winding and core state respectively, a multi-channel signal blind source separation algorithm based on the Sub-band decomposition independent component analysis combined with Gaussian mixed model (GMM) was proposed in this paper. Firstly, the winding and core voice print signals are separated based on the proposed multi-channel blind source separation algorithm. Secondly, the winding and core voice print signals of the transformer during normal operation were applied to construct the sample data sets respectively, which were used as training sets for winding and core GMM modeling. The obtained model was then used to diagnose abnormal signals of winding core vibration using normal signals mixed with different ratios of DC bias magnetism. In the test setup, it was experimentally demonstrated that the anomaly diagnosis algorithm has high sensitivity and effectiveness even in the presence of trace abnormal signal, enabling separate diagnosis of winding and core anomaly states. Finally, the abnormality diagnosis model is transplanted to an STM32-based intelligent voice print signal monitoring system to realize real-time diagnosis and visual display of the abnormal state of core and winding.