Transformer state analysis method and system based on deep learning
By using deep learning methods to automatically process transformer characteristic gas data, the problem of insufficient feature extraction capability in transformer condition analysis is solved, enabling accurate early fault identification and rapid response, thus meeting the intelligent operation and maintenance needs of power systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INNER MONGOLIA ELECTRIC POWER (GRP) CO LTD XILIN GOL ULTRA-HIGH VOLTAGE POWER SUPPLY BRANCH
- Filing Date
- 2026-04-24
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies lack the ability to process characteristic gas concentrations in depth and automatically in transformer condition analysis, which causes early and weak fault characteristics to be masked by noise, making it difficult to achieve accurate fault identification and rapid response, and failing to meet the intelligent and refined operation and maintenance needs of power systems.
A deep learning-based transformer state analysis method is adopted, which uses convolutional neural networks to automate the processing of characteristic gas data, including data preprocessing, convolutional layer feature extraction, pooling layer dimensionality reduction, fully connected layer combination, and backpropagation optimization, to form a trained state analysis model and realize the automated identification of transformer state.
It improves the accuracy and anti-interference capability of transformer condition identification, realizes the automation and efficiency of the analysis process, avoids missed judgments and misjudgments, and adapts to the intelligent operation and maintenance needs of diverse power production scenarios such as urban core substations.
Smart Images

Figure CN122432843A_ABST