| 引用本文: | 马富齐,穆睿昕,贾 嵘,等.基于声光融合成像特征解析的电力设备局部放电精细识别方法研究[J].电力系统保护与控制,2025,53(11):51-62.[点击复制] |
| MA Fuqi,MU Ruixin,JIA Rong,et al.Refined identification method for partial discharge in power equipment based on acoustic-optical fusion imaging feature analysis[J].Power System Protection and Control,2025,53(11):51-62[点击复制] |
|
| 摘要: |
| 局部放电是表征电力设备绝缘状态的重要指标，研究局部放电辨识对保障电力设备及电网安全运行意义重大。然而局部放电信号微弱，不同类型局部放电特征差异小，现有基于单数据源的局部放电监测方法信息利用率低、辨识精度有限。为此，提出了一种基于声光融合成像特征解析的电力设备局部放电精细识别方法。首先，对采集到的放电音频和声像图进行滑动特征提取，构成声光融合特征矩阵。其次，将特征矩阵嵌入多元时间序列，利用门控双轴编码模型并行地从时间轴方向和特征轴方向进行信息抽取、权重分配及特征重组。最后，计算重组特征向量属于各个类别的概率，实现局部放电高精度辨识。结果表明，所提方法能够实现对多种放电类型的精确识别，其准确率可达98.32%，相较基于单数据源特征的局部放电辨识表现出更好的检测效果。 |
| 关键词: 局部放电 声光融合成像 多元特征解析 时间序列 模式识别 |
| DOI：10.19783/j.cnki.pspc.241176 |
| 投稿时间：2024-09-02修订日期：2025-02-08 |
| 基金项目:陕西省自然科学基础研究计划(2024JC-YBQN- 0433)；国家科技部高端外国专家引进计划(G2023041010L) |
|
| Refined identification method for partial discharge in power equipment based on acoustic-optical fusion imaging feature analysis |
| MA Fuqi1,2,MU Ruixin1,JIA Rong1,WANG Bo2,ZHAO Yuhang1,MA Hengrui2 |
| (1. School of Electrical Engineering, Xi’an University of Technology, Xi’an 710054, China;
2. School of Electrical and Automation, Wuhan University, Wuhan 430072, China) |
| Abstract: |
| Partial discharge is an important indicator for assessing the insulation condition of power equipment, and accurate identification of partial discharge types is essential for ensuring the safe operation of both power equipment and power grid. However, due to weak partial discharge signals and the similar characteristics of difference types of partial discharges, existing partial discharge monitoring methods based on single data source suffer from low information utilization and limited identification accuracy. To address these challenges, a refined identification method for partial discharge in power equipment based on acoustic-optical fusion imaging feature analysis is proposed. First, sliding feature extraction is performed on the collected discharge audio and acoustic images to form a feature matrix of acoustic-optical fusion. The feature matrix is then embedded into a multivariate time series, and a gate controlled dual-axis encoding model is used to extract information, allocate weights, and recognize features in parallel along both the time and feature dimensions. Finally, the probability of the recognized feature vector belonging to each discharge category is calculated to achieve high-precision identification. Results show that the proposed method can achieve accurate identification of multiple types of discharge with an accuracy of up to 98.32%, outperforming identification methods based on single-source features. |
| Key words: partial discharge acoustic-optical fusion imaging multivariate feature analysis time series pattern recognition |
