Transformer partial discharge intelligent identification method based on deep learning

By combining electrical, acoustic, and thermal signal acquisition with operating parameters, and improving the deep learning model, the problems of high-frequency interference and dynamic operating condition adaptability in transformer partial discharge identification were solved, achieving high-precision fault identification and location.

CN122153803APending Publication Date: 2026-06-05UHVDC CENT OF STATE GRID SICHUAN ELECTRIC POWER CO
0 Cites 0 Cited by

Patent Information

Application Number
CN202610499716.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-16
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing deep learning-based transformer partial discharge identification technologies suffer from high-frequency interference, signal distortion, inability of the model to adapt to dynamic operating conditions, and lack of feature contribution feedback, resulting in low identification accuracy and difficulty in locating defects.

Method used

The model employs three-mode signal acquisition (electric, acoustic, and thermal) and combines it with operating parameters. It removes interference signals by fitting the mapping relationship between oil temperature and load rate, designs a cross-physical field feature fusion module, improves the ResNet-18 network, adds an operating condition adaptation layer, introduces a SHAP interpretable module, optimizes the model, and performs lightweight processing.

Benefits of technology

It improves the accuracy of partial discharge identification, adapts to dynamic operating conditions, provides feature contribution feedback, ensures that maintenance personnel can accurately locate defects, and improves maintenance efficiency.

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

Abstract

The invention belongs to the technical field of transformer partial discharge identification, and discloses a transformer partial discharge intelligent identification method based on deep learning, and the method comprises the steps: building an electro-acoustic-thermal three-mode collection architecture, capturing 1ns-level discharge pulses through a high-frequency current sensor, and enabling an ultrasonic sensor to be reasonably arranged on an oil tank wall, thereby avoiding the sound field superposition interference, and achieving the intelligent identification of the partial discharge of a transformer. The infrared thermal imager focuses on the easy-to-discharge area to collect thermal signals, and synchronously obtains working condition parameters such as load rate, oil temperature and the like; then, effective signal segments are reserved through a multi-modal signal cross-correlation matrix, interference signals with large deviation are removed, then electroacoustic-thermal characteristics are extracted through different networks, coupling characteristics are calculated in combination with the weight of a physical constraint layer and a cross-field interaction layer, and key characteristics are reserved; according to the method, the characteristic overlapping degree of air gaps and creeping discharge is effectively reduced, the influence of high-frequency interference and signal distortion is reduced, the partial discharge recognition precision is improved, time stretching and additive noise processing are combined, samples are expanded, and the proportion of faults to normal samples is reasonable.
Need to check novelty before this filing date? Find Prior Art