Converter transformer fault prediction method based on deep neural network

By constructing a model using deep neural networks and combining multi-scale convolutional neural networks, temporal convolutional networks, and attention mechanisms, the problem of balancing local mutations and long-term trends in converter transformer fault prediction was solved, achieving high sensitivity and consistency in fault prediction.

CN122455153APending Publication Date: 2026-07-24WUHAN UNIV OF TECH
0 Cites 0 Cited by

Patent Information

Application Number
CN202610532382.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-21
Publication Date
2026-07-24

AI Technical Summary

Technical Problem

Existing methods for predicting converter transformer faults struggle to balance local abrupt changes with long-term fault evolution trends, and the generated pseudo-samples violate gas diffusion and gas generation mechanisms, resulting in insufficient model generalization ability.

Method used

A deep neural network-based approach is adopted, which constructs a model using multi-scale convolutional neural networks, temporal convolutional networks, bidirectional long short-term memory networks, and attention mechanisms. The model is then combined with Mahalanobis distance and physical constraint rules to augment temporal samples for fault prediction.

Benefits of technology

It enables sensitive prediction of converter transformer faults, improves the sensitivity of fault prediction and the consistency of timing discrimination, and provides a robust fault early warning mechanism.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a converter transformer fault prediction method based on a deep neural network, and relates to the technical field of power equipment state monitoring, and the method comprises the steps: obtaining the concentration time sequence data of a plurality of dissolved gases in oil in a historical time period of a converter transformer, and carrying out the preprocessing; mapping the data into a supervised learning sample set by adopting a fixed-length sliding window mechanism, and dividing the supervised learning sample set into a training set and a test set; aiming at minority class samples containing fault samples and fault precursor samples in the training set, adopting a multivariate dynamic time warping method based on mahalanobis distance to carry out time sequence sample enhancement, and screening to obtain an enhanced training set; constructing and training a multi-classification deep neural network model based on the enhanced training set, and outputting single-window prediction probabilities of various faults; and inputting real-time multi-in-oil dissolved gas concentration time sequence data subjected to data preprocessing and window division consistent with the training stage into the trained multi-classification deep neural network model, and outputting a final fault type.
Need to check novelty before this filing date? Find Prior Art