| 引用本文: | 徐敏锐,李云,卢树峰,窦晓波,陈刚,郭家豪.基于D-S证据组合规则的双模型融合局部放电模式识别方法[J].电力自动化设备,2021,41(11): |
| XU Minrui,LI Yun,LU Shufeng,DOU Xiaobo,CHEN Gang,GUO Jiahao.Recognition method of partial discharge pattern based on double models fusion with D-S evidence combination rule[J].Electric Power Automation Equipment,2021,41(11): |
|
| 摘要: |
| 针对传统的局部放电模式识别存在的特征提取单一、识别准确率低等缺点，提出了一种基于D-S证据组合规则的双模型融合局部放电模式识别方法。根据基于相位信息的局部放电(PRPD)谱图的统计数据特征和图像特征的特点，分别建立了反向传播(BP)识别模型和卷积神经网络(CNN)识别模型。根据2个识别模型的识别结果，提出了基于信息熵改进的D-S证据组合规则以解决常见的悖论问题，基于此建立了判定模型，更好地融合了2个识别模型的输出结果，实现了2种特征识别的优势互补。根据实际数据测试，与单一模型对比，所提方法可以稳定、准确地识别局部放电模式。 |
| 关键词: 局部放电 特征提取 机器学习 模式识别 D-S证据组合 |
| DOI：10.16081/j.epae.202106002 |
| 分类号:TM507 |
| 基金项目:国家电网公司科技项目(5600-201918181A-0-0-00) |
|
| Recognition method of partial discharge pattern based on double models fusion with D-S evidence combination rule |
|
XU Minrui1, LI Yun2, LU Shufeng1, DOU Xiaobo3, CHEN Gang1, GUO Jiahao1
|
|
1.Marketing Service Center, State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 210096, China;2.School of Cyber Science and Engineering, Southeast University, Nanjing 211189, China;3.School of Electrical Engineering, Southeast University, Nanjing 210096, China
|
| Abstract: |
| The traditional PD(Partial Discharge) pattern recognition has the disadvantages of single feature extraction and low recognition accuracy, aiming at which, a PD pattern recognition based on double models fusion with D-S evidence combination rule is proposed. According to the characteristics of the statistical data and image features of PRPD(Partial Discharge based on Phase Information) spectrum data, BP(Back Propagation) recognition model and CNN(Convolutional Neural Network) recognition model are respectively established. According to the output results of the two recognition models, the improved D-S evidence combination rule based on information entropy is proposed to solve the common paradox problems. The decision model is established based on the improved D-S evidence combination rule, by which, the output results of the two recognition models are better integrated and complement the advantages of the two feature recognition methods. The actual data test shows that compared with the single model, the proposed method can recognize the PD patterns stably and accurately. |
| Key words: partial discharge feature extraction machine learning pattern recognition D-S evidence combination |
