Research on Deformation Fault Diagnosis of Transformer Windings Based on a Highly Sensitive Multimodal Feature System

The current mainstream methods for online detection of transformers all have shortcomings such as low sensitivity and susceptibility to interference from the testing environment. Aiming at the shortcomings of the existing online detection methods for transformer winding deformation in terms of feature sensitivity and diagnostic accuracy, this paper proposes a fault intelligent diagnosis method based on high sensitivity multimodal feature fusion. First, the winding deformation experiment is designed for typical fault data, which is obtained to extract multiple frequency and time domain response features and construct a multidimensional feature library. Subsequently, principal component analysis is used to evaluate the sensitivity of each feature to different faults and establish a highly sensitive multimodal feature system. On this basis, a TCN-BiGRU-PHA diagnostic model combining time convolutional network, bidirectional gated loop unit and attention mechanism is constructed to realize accurate identification of winding deformation faults. The experimental results show that the method has higher recognition accuracy under multiple types of faults, which provides feasible ideas and methodological support for realizing online intelligent monitoring of transformer winding deformation.