Transformer zero-sample fault diagnosis based on multimodal data fusion

As the core equipment of power systems, the accurate diagnosis and location of faults in transformers has become a pressing problem in the transformer field. In this paper, the concept of multimodality and zero-sample is integrated into transformer fault diagnosis, and a multimodal zero-sample transformer fault diagnosis model is developed to address the high coupling between transformer faults, the difficulty of collecting some faults and the lack of sample data. The model is based on the semantic embedding of multidimensional information into the attribute feature space to achieve multimodal fusion, which in turn enables a flexible global online supervision mechanism for transformer fault diagnosis system; secondly, considering the problem of missing data and fault location deviation for specific types of faults in transformers, a transformer knowledge map is constructed and a fault attribute matrix is deduced based on expert experience, with the help of its attribute features and fault instance The model is generalised to learn from its attribute features and fault instance classification. The proposed method enables the identification of zero-sample faults and fault-specific localisation in the context of multimodal fusion, with high system flexibility, wide fault coverage and good diagnostic results.