Abstract: In order to effectively detect the transformer DC magnetic bias state, this paper starts from the perspective of multi-channel information fusion, and proposes a new method for identifying the bias state of a transformer based on symmetrized dot pattern (SDP) and 2-dimensionlal local nonngeative matrix factorzization(2DLNMF). Firstly, SDP algorithm is used to fuse the multi-channel vibration signals into SDP image features. Then, the 2DLNMF algorithm was used to optimize its dimensionality reduction, according to which the transformer magnetic bias state recognition model based on the support vector machine (SVM) algorithm was built. The research results show that information fusion method based on the SDP-2DLNMF fully shows the characteristics of the differences between different signal, the obtained low-dimensional characteristics can effectively reflect the degree of DC magnetic bias of the transformer, and the SVM state recognition model established on this basis has high recognition accuracy, which provides technical support for transformer state monitoring.

    Key words: transformer; DC magnetic bias; symmetrized dot pattern; 2-dimensionlal local nonngeative matrix factorzization; support vector machine