Transformer Data Correction Approaches Based on GMM and BiGRU

Given that the presence of irregular data, such as absent values, noise and erroneous values, in transformer monitoring data leads to a reduction in the accuracy of transformer condition prediction and assessment, a data cleansing method based on a Gaussian mixture model (GMM) and a bi-directional gated recurrent unit (BiGRU) is proposed. Initially, the original data are clustered and analyzed using the GMM data clustering algorithm in order to differentiate between standard and anomalous data. Subsequently, the data cleaning rules are proposed, distinguishing between data that is “dirty data” and abnormal transformer state data, based on the identified outliers. Subsequently, the “dirty data” identified by the GMM are subjected to cleansing by the BiGRU. The effectiveness of the proposed method was evaluated using actual transformer monitoring data, demonstrating that it can accurately identify and effectively remove “dirty data”.