| 基于改进Cascade R-CNN网络的电力变压器局部放电模式识别 |
投稿时间：2021-10-21 修订日期：2021-11-26  点此下载全文 |
| 引用本文： |
| 摘要点击次数: 106 |
| 全文下载次数: 0 |
|
| 基金项目:南方电网重点科技项目 |
|
| 中文摘要:电力变压器的安全性影响着电力系统的稳定，且局部放电是变压器绝缘老化的重要原因。为了保障电力系统能够安全稳定地运行，需要对变压器的局部放电进行在线监测。电力变压器常见的局部放电主要为气隙放电、针板放电、悬浮放电和沿面放电。本文以Cascade R-CNN为基础，在Backbone中插入可变形卷积学习几何变换能力，并设计了常规Cascade R-CNN实验作为对照以验证模型的有效性，最后训练结果证明引入可变性卷积的改进能够较好地完成识别任务。此外，以不同样本数作为训练集，表明改进Cascade R-CNN网络有着更大的应用潜力，在更大的样本支撑下能够继续提升准确率。 |
| 中文关键词:电力变压器 局部放电 Cascade R-CNN 模式识别 |
| |
| Partial Discharge Pattern Recognition of Power Transformer Based on Improved Cascade R-CNN Network |
|
|
| Abstract:The safety of power transformers affects the stability of power systems, and partial discharge is an important cause of transformer insulation aging. In order to ensure the safe and stable operation of the power system, online monitoring of the partial discharge of the transformer is required. The common partial discharges of power transformers are mainly air gap discharge, needle plate discharge, floating discharge and creeping discharge. Based on Cascade R-CNN, this paper inserts deformable convolution into Backbone to learn geometric transformation capabilities. And designed the normal Cascade R-CNN experiment as a control to verify the effectiveness of the model, and finally the training results show that the improved variable convolution can better complete the recognition task. In addition, using different sample numbers as training sets indicates that the improved Cascade R-CNN network has greater application potential and can continue to improve accuracy with the support of larger samples. |
| keywords:Power transformer Partial discharge Cascade R-CNN Pattern recognition |
| 查看全文 查看/发表评论 下载pdf阅读器 |
位访问者,今日访问