基于斯托克斯向量的MOCT故障识别模型的研究

针对传统电磁电流互感器故障时因磁饱和造成继保误动或拒动现象,提出了一种磁光电流互感器（M0CT）快速识别故障发生及类型的方法.该方法以通过石榴石型M0CT的偏振光的三个斯托克斯分量作为特征量,建立支持向量机识别模型,模型核参数及惩罚参数由粒子群寻优选取.此建模方法仅与斯托克斯分量及故障电流等级有关,与线性双折射及环境变量无关,避免了温度等外界因素对识别的干扰.实验结果表明,识别模型受温度影响小,当训练完成的识别模型工作于不同的温度条件下时,识别误差保持在2%以内.运行中的M0CT可以依靠斯托克斯分量及该模型准确识别故障电流等级,迅速判别出故障类型,精度较其他识别方法更高....

Full description

Saved in:
Bibliographic Details
Published in计算机应用研究 Vol. 34; no. 5; pp. 1358 - 1361
Main Author 刘宇宽 马立新 何亮 张建宇
Format Journal Article
LanguageChinese
Published 上海理工大学光电与计算机工程学院,上海,200093%上海理工大学光电与计算机工程学院,上海200093 2017
上海市现代光学系统重点实验室,上海200093
Subjects
偏振性质
支持向量机
故障识别
永磁薄膜
磁光电流互感器
粒子群优化
支持向量机
故障识别
粒子群优化
偏振性质
magnetic film
永磁薄膜
optical polarization
SVM
磁光电流互感器
PSO
magnetic optical current transformer(MOCT)
fault identification
Online AccessGet full text
ISSN1001-3695
DOI10.3969/j.issn.1001-3695.2017.05.018

Cover

More Information
Summary:针对传统电磁电流互感器故障时因磁饱和造成继保误动或拒动现象,提出了一种磁光电流互感器（M0CT）快速识别故障发生及类型的方法.该方法以通过石榴石型M0CT的偏振光的三个斯托克斯分量作为特征量,建立支持向量机识别模型,模型核参数及惩罚参数由粒子群寻优选取.此建模方法仅与斯托克斯分量及故障电流等级有关,与线性双折射及环境变量无关,避免了温度等外界因素对识别的干扰.实验结果表明,识别模型受温度影响小,当训练完成的识别模型工作于不同的温度条件下时,识别误差保持在2%以内.运行中的M0CT可以依靠斯托克斯分量及该模型准确识别故障电流等级,迅速判别出故障类型,精度较其他识别方法更高.
Bibliography:51-1196/TP
Due to the phenomenon that traditional electromagnetic current transformer encountesoperation of the relay protection when it appeared magnetic saturation under fai lure conditions, this paper proposed a methodof the magnetic optical current transformer （ M0CT） for fastly identifying fault. In order to establimodel, the method took three Stokes components of polarized l ight which passed through garnet optical racteristics. It optimized the nuclear and punishment parameters by particle swarm optimization. The model was not associatedwith l inear birefringence or environment variables but was associated with Stokes component and fault current grade, so it avoi-ded the interference of external factors. The result shows that temperature has little effect on the identification model. Although the model works at different temperature conditions, the precision of trained recognition model keep inside of 2%. M0CT hashigher accuracy than other identification methods. It can accurately identify the fault cur
ISSN:1001-3695
DOI:10.3969/j.issn.1001-3695.2017.05.018