Pattern Recognition Method of Transformer Multi-source Partial Discharge Based on Parallel Eigen Domain

Abstract

A transformer multi-source partial discharge pattern recognition method based on parallel feature domains, extracting multi-source partial discharge signals to construct multi-source partial discharge time-domain signal sets and time-frequency domain signal sets; combining multiple single neural network autoencoders into two stacks Encoder. The sigmoid function is selected as the activation function between the stacked encoder network layers, and the activation function is used to obtain the activation value of the next network layer. Add a regularization term to the loss function that adjusts the network parameters of each layer in the stacked encoder. The optimal solution method for setting the loss function is the proximal-oriented stochastic subgradient algorithm. A softmax is added as a classification layer of the neural network; a stacked encoder is used for parallel training on multi-source partial discharge time-domain signal sets and time-frequency domain signal sets. The feature matrix data has corresponding labels, and the network parameters are fine-tuned by comparing the classification results. This method has the advantages of high classification accuracy and strong generalization ability of the deep learning model, and is suitable for multi-source partial discharge pattern recognition of transformers and other occasions.

Description

technical field [0001] The invention relates to the field of transformer partial discharge pattern recognition, in particular to a transformer multi-source partial discharge pattern recognition method based on parallel feature domains. Background technique [0002] The traditional multi-source partial discharge identification method is based on the structural feature quantity, but although the artificially constructed features have clear physical meaning, but the dimension is lower; and the deep learning algorithm uses the nonlinear mapping between the hidden layers to integrate the multi-source partial discharge The discharge feature information is abstracted into a multi-dimensional matrix, which covers more feature information and can better represent the original multi-source partial discharge information. [0003] As deep learning is one of the most efficient feature extraction tools, as the number of network layers increases, redundant parameters between network layers...

AI Technical Summary

Problems solved by technology

[0005] ①. The signal separation effect is poor; ②. The artificially extracted signal features are not accurate enough; ③. The neural network algorithm used for pattern recognition has problems such as overfitting during training

Images