变压器表面振动信号基频幅值分析与预测
发布时间:2019-01-20 17:47
【摘要】:变压器是电力系统关键设备之一,其运行状况对电网的安全、稳定运行具有重要影响。变压器表面振动信号中包含着丰富的变压器状态信息,国内外学者对基于振动分析的变压器在线监测和故障诊断技术做了大量研究,并取得了许多研究成果。变压器振动基频(100Hz)幅值大小是分析和评判变压器运行状态和故障诊断的重要依据,但由于多种因素的影响,理论分析正常运行中变压器表面振动的基频幅值困难诸多,尚没有成熟的基于基频幅值的变压器状态监测方法。本文针对变压器表面振动信号采集和分析实际需求,结合已有研究结果,分析探讨了传感器的选型、振动测点的选择以及采集参数等问题,设计实现了一套便携式振动信号采集系统,完成了多台次运行中变压器表面振动数据的采集;对变压器表面振动实测数据进行了频域分析和能量分析,并结合运行电压和负载电流数据,分析了振动信号基频幅值与运行工况的关系,结果表明基频幅值大小受多重因素复杂影响,实测值与理论计算值差异显著;本文给出一种基于广义回归神经网络(GRNN)的基频幅值预测方法,用于正常运行状态下的变压器表面振动基频幅值预测。根据变压器运行电压、负载电流、油温等运行工况数据以及表面振动历史数据进行网络训练,训练后的网络可根据实时运行数据预测变压器表面振动基频幅值。运行中变压器表面振动实测信号分析表明,本文方法比原有方法预测精度高,可为基于振动的变压器在线监测提供参考。最后本文给出一种典型样本筛选方法,首先基于模糊熵理论计算特征权重,然后根据运行工况数据间的加权欧氏距离对训练样本进行筛选,实测数据分析表明该方法可以显著压缩训练数据,降低数据冗余,提高网络训练速度和计算速度。
[Abstract]:Transformer is one of the key equipments in power system. The vibration signal of transformer surface contains abundant transformer state information. Many researches have been done on on-line monitoring and fault diagnosis of transformer based on vibration analysis, and many research results have been achieved. The amplitude of transformer vibration base frequency (100Hz) is an important basis for analyzing and judging transformer operation state and fault diagnosis. However, due to the influence of many factors, it is difficult to analyze the fundamental frequency amplitude of transformer surface vibration in normal operation. There is no mature transformer condition monitoring method based on fundamental frequency amplitude. In this paper, according to the actual demand of vibration signal acquisition and analysis on transformer surface, combined with the existing research results, the selection of sensor, the selection of vibration measuring points and the acquisition parameters are analyzed and discussed. A portable vibration signal acquisition system is designed and implemented. The frequency domain analysis and energy analysis of the measured data of transformer surface vibration are carried out, and the relationship between the fundamental frequency amplitude of vibration signal and the operating condition is analyzed by combining the data of operating voltage and load current. The results show that the amplitude of the fundamental frequency is affected by many complex factors, and there is a significant difference between the measured value and the theoretical value. In this paper, a prediction method of fundamental frequency amplitude based on generalized regression neural network (GRNN) is presented, which can be used to predict the fundamental frequency amplitude of transformer surface vibration under normal operation. Network training is carried out according to the operating condition data of transformer operating voltage, load current, oil temperature and history data of surface vibration. The trained network can predict the fundamental frequency amplitude of transformer surface vibration based on real-time operation data. The analysis of the measured signals of transformer surface vibration in operation shows that the proposed method is more accurate than the original method and can be used as a reference for on-line monitoring of transformers based on vibration. Finally, a typical sample selection method is presented. Firstly, the feature weights are calculated based on the fuzzy entropy theory, and then the training samples are screened according to the weighted Euclidean distance between the operating condition data. The analysis of measured data shows that this method can significantly compress the training data, reduce the data redundancy, and improve the training speed and computing speed of the network.
【学位授予单位】:华北电力大学
【学位级别】:硕士
【学位授予年份】:2017
【分类号】:TM41
[Abstract]:Transformer is one of the key equipments in power system. The vibration signal of transformer surface contains abundant transformer state information. Many researches have been done on on-line monitoring and fault diagnosis of transformer based on vibration analysis, and many research results have been achieved. The amplitude of transformer vibration base frequency (100Hz) is an important basis for analyzing and judging transformer operation state and fault diagnosis. However, due to the influence of many factors, it is difficult to analyze the fundamental frequency amplitude of transformer surface vibration in normal operation. There is no mature transformer condition monitoring method based on fundamental frequency amplitude. In this paper, according to the actual demand of vibration signal acquisition and analysis on transformer surface, combined with the existing research results, the selection of sensor, the selection of vibration measuring points and the acquisition parameters are analyzed and discussed. A portable vibration signal acquisition system is designed and implemented. The frequency domain analysis and energy analysis of the measured data of transformer surface vibration are carried out, and the relationship between the fundamental frequency amplitude of vibration signal and the operating condition is analyzed by combining the data of operating voltage and load current. The results show that the amplitude of the fundamental frequency is affected by many complex factors, and there is a significant difference between the measured value and the theoretical value. In this paper, a prediction method of fundamental frequency amplitude based on generalized regression neural network (GRNN) is presented, which can be used to predict the fundamental frequency amplitude of transformer surface vibration under normal operation. Network training is carried out according to the operating condition data of transformer operating voltage, load current, oil temperature and history data of surface vibration. The trained network can predict the fundamental frequency amplitude of transformer surface vibration based on real-time operation data. The analysis of the measured signals of transformer surface vibration in operation shows that the proposed method is more accurate than the original method and can be used as a reference for on-line monitoring of transformers based on vibration. Finally, a typical sample selection method is presented. Firstly, the feature weights are calculated based on the fuzzy entropy theory, and then the training samples are screened according to the weighted Euclidean distance between the operating condition data. The analysis of measured data shows that this method can significantly compress the training data, reduce the data redundancy, and improve the training speed and computing speed of the network.
【学位授予单位】:华北电力大学
【学位级别】:硕士
【学位授予年份】:2017
【分类号】:TM41
【参考文献】
相关期刊论文 前10条
1 王广庭;李阳海;杨涛;苏世玮;赵家毅;;基于振动信号分析的变压器故障诊断研究进展[J];噪声与振动控制;2016年05期
2 陈伟根;万福;顾朝亮;邹经鑫;漆薇;王品一;;变压器油中溶解气体拉曼剖析及定量检测优化研究[J];电工技术学报;2016年02期
3 周科峰;叶婷;;南京电网变压器油色谱在线监测的应用探讨[J];电工技术;2016年01期
4 王春宁;耿志慧;马宏忠;金基平;李凯;;基于振动的电力变压器铁心松动故障诊断研究[J];高压电器;2015年12期
5 陈东超;徐婧;洪瑞新;顾煜炯;何成兵;;基于广义回归神经网络的旋转机械振动特征预测[J];汽轮机技术;2015年05期
6 马宏忠;周宇;李凯;许洪华;王涛云;;基于振动的变压器绕组压紧状态评估方法[J];电力系统自动化;2015年18期
7 朱叶叶;汲胜昌;张凡;刘勇;董鸿魁;崔志刚;吴佳玮;;电力变压器振动产生机理及影响因素研究[J];西安交通大学学报;2015年06期
8 林U,
本文编号:2412247
本文链接:https://www.wllwen.com/kejilunwen/dianlidianqilunwen/2412247.html