《毕业论文外文翻译-利用人工神经网络和绕组的传递函数定位变压器的匝间短路故障.doc》由会员分享，可在线阅读，更多相关《毕业论文外文翻译-利用人工神经网络和绕组的传递函数定位变压器的匝间短路故障.doc（19页珍藏版）》请在taowenge.com淘文阁网|工程机械CAD图纸|机械工程制图|CAD装配图下载|SolidWorks_CaTia_CAD_UG_PROE_设计图分享下载上搜索。
1、 中文翻译 利用人工神经网络和绕组的传递函数定位变压器的匝间短路故障 Mohsen Faridi Ebrahim Rahimpour摘要：为了用自动化的程序定位在变压器绕组匝间故障,提出了一种叫人工神经网络(ANN)的方法。为此,我们定制了一个特制的配电变压器，通过测试其绕组来验证该方法的可靠性。这使这个实验设计合理，决定用任意两个相邻短路故障作为研究对象。利用低电压脉冲(LVI)测试绕组在正常和故障的条件的频率响应。从频率响应中提取特征数据用于检验和测试人工神经网络ANN。结果表明,该方法能够准确在绕组匝间故障的位置。关键词：匝间短路,变压器绕组;传递函数;神经网络。 变压器被认为是电网中最
2、重要的设备之一。变压器故障降低电力网络的可靠性,同时也对变压器的运行造成灾害性的损坏。所以提高可靠性和诊断其故障将有效减少支损失。为了在早期阶段发现变压器匝间短路故障，有必要开发合适的方法来检测它们。在这之前，虽然有采用了很多方法,但显然还没有出现一种可靠的方法来检测它们的故障。 导言 配电变压器的匝间故障是一种常见的故障。到目前为止,已经有很多不同的方法被用来检测这种故障。包括传递函数计算，1 - 2,3有限元分析,小波分析4,变压器电流序列分析，56和漏磁通密度计算。在本文中,提出的神经网络(ANN)是一个基于传递函数的测量来检测匝间故障。为此,定制一个特制配电变压器，变压器的绕组是人工模
3、拟匝间短路故障。然后测量故障之前和故障之后的传递函数，并用人工神经网络ANN定位匝间短路位置。 问题分析 在配电变压器匝间故障是主要是由不规范操作和绕组热点温升过高引起。这些可能导致导体的绝缘损坏。因此在绕组匝间短路的故障主要可能是在高压绕组的外层或者高压绕组（HV）和低压绕组（LV）的顶部或者低部。 在配电变压器匝间故障定位的传统方法是应用一个测试电压绕组,然后逐渐增加其大小,直到绝缘缺陷的处变得碳化。然后,缺陷位置可以观察到。但是这种方法是破坏性的,并可能导致损害相邻匝。因此有必要找到无损定位匝间短路故障的方法。 根据绕组匝间短路故障的模型参数的影响和传递函数的测量和分析来诊断匝间短路时合
4、理的。在这个研究上电压转移和转移导纳函数是用来定位匝间短路故障。 测试绕组的特性 在这个工作电压转移和转移导纳函数是用来定位匝间短路故障 图1显示了专门准备的绕组20/0.4KV,原理图50 kva变压器用于进行实验测试。层类型高压绕组由15层。到第31个接头，更有可能变成匝间短路故障。因为短路选取两个接头产生匝间短路缺陷。考虑到选取的邻近两个匝间的距离,在查查寻过程中找到84匝人工匝间短路模型。 图1检测电路和测量系统 两个不同的频率响应测量的目的故障定位。图2显示了测量的测试电路转移导纳函数。在这个电路低压(LV)绕组短路的和中性的高压绕组接地通过50阻力。一连串的低压脉冲用于激励高压侧线
5、圈。高压侧电压及其中性电流测量在时间域分别作为输入和输出信号。利用傅里叶变换计算测量信号的导纳来得出传递函数。 Io和Vi分别是接地电流和外加电压。 图2.测试导纳的电路图 图3.测量转移电压的电路图图4.试验设备 在图3的测试电路测量电压传递函数。在这个电路中,接线柱的高压和低压绕组接地直接和脉冲应用于高压终端。电压转移函数计算使用的傅里叶变换测量高压和低压电压:V0和Vi是对低压绕组LV的电压传输和高压绕组的应用。应用低压脉冲有10个ns倍上升。输入和输出信号使用数字示波器测量500 msam /秒和105样本/记录。完整的测试电路安排图4所示。在图5中,转移导纳的测量输入和输出信号和传输
6、电压测试电路圈所示的正常状况。测量信号在测试电路第2和第4匝，第26匝和30匝的短路转身是短路 如图6所示。计算传递函数图6所示的信号提出了图7到10。 图，5正常情况下的绕组的信号输出 测量第2和第4之间还有第26和第30匝之间的短路故障情况下的输入和输出信号 图7. 测量在正常运行时第26和第30匝之间的短路故障的传递导纳 图8. 测量在正常运行时第2和第4匝之间的短路故障的传递导纳 图9. 绕组的传递电压在正常情况下和第26 和第30匝之间的短路故障。 图10. 绕组的传递电压在正常情况下和第2 和第4匝之间的短路故障。 人工神经网络智能算法可以学习之后一系列的训练模式和分类出新模式。为
7、了缺陷定位，两个相似的人工神经网络的具有反向学习算法得到应用。16主导极点的振幅再计算电压转移和转移导纳函数作为这两个网络作为输入。模拟匝间故障的结果,被分为两个组,一组用于训练网络,另一组用于测试。这样,42匝间故障信息被用来测试ANN和其他42个错误被用来测试它。所以,在训练阶段,两个4216和421矩阵每个网络的应用作为输入和预期的输出。用试错法,人工神经网络结构包括16，4和1神经元的输入,隐层和输出层,分别独立的。得出人功智能网络（ANN）故障能准确合理的定位匝间短路故障。 图11.用ANN神经网络测试转移电压的结果 图12.用ANN神经网络测试传递导纳的结果 用人工神经网络定位 图
8、11和图12显示的输出ANN的输入训练数据集和测试数据集分别传输电压和传输导纳函数。可以看到,ANN学习模式和也能成功定位故障。 最大、最小和平均误差测试数据传输电压输入1.949,0.0065,0。分别为825。这些值在应用的情况传输导纳函数的输入是2.352,0.2502,0。分别为7302。也在这两种情况下,训练网络能够准确定位缺陷在95%缺陷位置的测试数据集。 结论 在这篇文章中,提出了基于神经网络来定位匝间短路故障得方法。这个网络使用特征值法提取传输电压和传输导纳函数的数据。利用人工模拟变压器匝间短路故障的试验数据来训练神经网络的定位故障。结果由此证明利用神经网络定位匝间的可行性。
9、Localization of Turn-to-Turn Fault in TransformersUsing Artificial Neural Networks and WindingTransfer Function Mohsen FaridiIslamic Azad University, Khodabandeh Branch Khodabandeh, IranEbrahim RahimpourABB AG, Power Transformers Bad Honnef, GermanyAbstractTo automate the procedure of localizing tur
10、n-to-turn faults in transformers windings, a method is proposed by employing of Artificial Neural Networks (ANN). For this purpose, a specially made distribution transformer winding is used as a test object to approve the capability of proposed method. This winding is appropriately designed to perfo
11、rm short circuit faults between any two desired adjacent turns. Then the frequency response of winding in both healthy and faulty conditions is measured using the Low Voltage Impulse (LVI) test. Extracted features from frequency responses are used to train and test the proposed ANN. The results show
12、 that this method is able to determine the location of turn-to-turn fault in winding.Keywords-component;Turn-to-TurnFault;TransformerWinding; Transfer Function; Neural NetworkI INTRODUCTIONTransformers are supposed to be one of the most important equipments in power networks. Transformer failures no
13、t only reduce these networks reliability, but also cause catastrophic damages to their active parts. So any efforts to increase their reliability and diagnosing their faults would effectively reduce expenditures. Regarding to importance of recognition and detection of transformer internal faults in
14、their early stages of appearance, it is necessary to develop suitable methods to detect them. Many works already have been performed in this context before, but clearly it has not been introduced any reliable method to detect their faults yet.Turn-to-turn fault is a common cause of distribution tran
15、sformers failures. Up to now, many different methods have been used to detect this fault such as transfer function calculation 1-2, finite element analysis 3, wavelet analysis 4, transformer current sequence analysis 5 and leakage factor calculation method 6. In this paper, an Artificial Neural Netw
16、ork (ANN) based on transfer function measurement is proposed to detect turn-to-turn faults. For this purpose, a special made winding of a distribution transformer is manufactured to simulate artificial turn-to-turn faults. Thendifferent transfer functions of this winding are measured before and afte
17、r implementing defects and then their extracted features applied to ANN to localize defect site.II. PROBLEM DEFINITIONTurn-to-turn faults in distribution transformers are caused mainly by careless transportation and excessive temperature rise in hot spots of windings. These might lead to damages to
18、conductors insulation. Therefore the main probable sites for turn-to-turn faults in windings are those turns which are located in outer layers of High Voltage (HV) winding or in top or bottom of both Low Voltage (LV) and HV windings.The conventional method for turn-to-turn fault localization in dist
19、ribution transformers is to apply a test voltage to their windings and then increase its magnitude gradually until the insulation of defect site becomes carbonized. Afterwards, defect location could be discriminated by visual inspections. But this method is destructive and may result in damages to a
20、djacent turns. Therefore it is favorable to find a non-destructive method to localize such defects.Due to the affect of turn-to-turn faults in windings model parameters, transfer function measurement and analysis is supposed to be suitable for their detection and localization. In this work the trans
21、fer voltage and the transfer admittance functions are employed to localize turn-to-turn fault.III. TEST WINDING CHARACHTERISTICSFig. 1 shows the schematic diagram of specially prepared winding of 20/0.4KV, 50KVA transformer which is used to perform experimental tests. The layer type HV winding consi
22、sts of 15 layers. Up to 31 joints are extracted from those turns which are more likely to be subjected to turn-to-turn faults. By short circuiting two extracted joints turn-to-turn defects are generated. Considering the proximity of those turnswhich are sampled out, totally 84 states of artificial t
23、urn-to-turn defects are simulated in this research work.Figure 1.Schematic diagram of test winding turs and joints.IV.TEST CIRCUITS AND MEASURING SYSTEMTwo different frequency responses are measured for the aim of fault localization. Fig. 2 shows the test circuit for measuring transfer admittance fu
24、nction. In this circuit the Low Voltage (LV) winding is short circuited and the neutral of HV winding is grounded via a 50 resistance. A train of low voltage impulses is applied to HV terminal to excite the winding. The HV terminal voltage and its neutral currents are measured in time domain as inpu
25、t and output signals respectively. The admittance transfer function is calculated using Fourier Transforms of measured signals as follows:TFAdmitt anceFFT ( Io )FFT (Vi )Where Io and Vi are the ground current of HV winding and applied voltage respectively.In Fig. 3 the test circuit for measuring the
26、 transfer voltage function is shown. In this circuit, the neutral terminal of both HV and LV windings is grounded directly and the impulse is applied to HV terminal. The transfer voltage function is calculated using the Fourier Transforms of measured HV and LV voltages:TFTransferVoltageFFT (Vo )FFT
27、(Vi )Where Vo and Vi are the transferred voltage to LV winding and applied voltage to HV winding respectively.JJJ. TEST WINDING CHARACHTERISTICSFig. 1 shows the schematic diagram of specially prepared winding of 20/0.4KV, 50KVA transformer which is used to perform experimental tests. The layer type
28、HV winding consists of 15 layers. Up to 31 joints are extracted from those turns which are more likely to be subjected to turn-to-turn faults. By short circuiting two extracted joints turn-to-turn defects are generated. Considering the proximity of those turns Figure 2.Test circuit diagram for trans
29、fer admittance function measurement. Figure 3.Test circuit diagram for transfer voltage function measurement.The applied low voltage impulses have 10ns rise times. Input and output signals are measured using a digital oscilloscope with 500Msam/sec and 105 samples per record. The complete test circui
30、t arrangement is shown in Fig. 4.V. MEASUREMENTS RESULTSIn Fig. 5, the measured input and output signals of both transfer admittance and transfer voltage test circuits in healthy condition of winding are shown. The measured signals in both test circuits when the 2nd and 4th turns as well as 26th and
31、 30th turns are short circuited are depicted in Fig. 6. The calculated transfer functions of signal which are shown in Fig. 6 are presented in Fig. 7 through 10. Figure 5.Measured input and ouput signals from winding healthy condition. Figure 6. Measured input and ouput signals when 2nd and 4th turn
32、s as well as 26th and 30th turns are short circuitted. Figure 7. Winding transfer admittance functions in healty condition and when 26th and 30th turns short circuitted. Figure 8. Winding transfer admittance functions in healty condition and when 2nd and 4th turns short circuitted. Figure 9. Winding
33、 transfer voltage functions in healty condition and when 26th and 30th turns short circuitted.VI.DEFECT LOCALIZATION USING ANNANNs are intelligent algorithms which could learn a set of training patterns and afterwards classify new patterns. For defect localization two similar ANNs with back propagat
34、ion learning algorithm are used. The amplitudes of 16 dominant poles of calculated transfer voltage and transfer admittance functions are applied to these two networks as inputs. The results obtained from simulated turn-to-turn faults, were divided to two separate groups, one for training the networ
35、ks and the other for testing them. In this way, 42 turn-to-turn faults information were used to train ANN and the other 42 faults were used to test it. So, in training phase, two 4216 and 421 matrices were applied as input and desired outputs of each network. Using a try and error method, ANNs struc
36、tures which include 16, 4 and 1 neurons in input, hidden and output layers respectively, found to be suitable for defect localization.Fig. 11 and Fig. 12 show the output of ANN for training data set of inputs as well as test data set for transfer voltage and transfer admittance functions respectivel
37、y. As can be seen, the ANN was capable to learn the patterns and could successfully localize the site of defects too.The maximum, minimum and average error for test data in case of transfer voltage inputs are 1.949, 0.0065 and 0.825 respectively. These values in the case of applying inputs of transf
38、er admittance function are 2.352, 0.2502 and 0.7302 respectively. Also in both cases the trained networks were able to localize defects accurately in 95% defect locations of test data set Figure 11. ANN test results for transfer voltage functions. Figure 12. ANN test results for transfer admittance
39、functions.VII. CONCLUSIONIn this paper, a turn-to-turn fault localization method basedon neural network was proposed. This network uses the features extracted from both transfer voltage and transferadmittance functions. The feasibility of fault localization oftrained ANN was tested using the experim
40、ental data whichwere obtained from measurements on artificial defectssimulated on a specially manufactured winding. Resultsapprove the ability of proposed ANN in this context.ACKNOWLEDGMENTThe authors would like to thank Iran-Transfo Co. for itssupports and collaborations in tests and measurements.R
41、EFERENCES1 E. Rahimpour, and D. Gorzin, “A New Method For Comparing theTransfer Function of Transformers in order to Detect the Location andAmount of Winding Faults. ”, Electrical Engineering, Vol.88, 2006, pp.411-416.2 E. Rahimpour, J. Christian, , K. Feser, and H. Mohseni, “Modellierungder Transfo
42、rmatorwicklung zur Berechnung der bertragungs funktionfr die Diagnose von Transformatoren. ”, Elektrie, No.54/1-2, 2000, pp.18-30.3 H. Wang, and K. L. Bulter, “Finite Element Analysis of InternalWinding Faults in Distribution Transformers”, IEEE Trans. on PowerDel., Vol.16, No.3, 2001, pp. 422-428.4
43、 M. R. Rao, and B. P. Singh, “Detection and Localization of InterturnFault in the HV Winding of a Power Transformer Using Wavelets. ”IEEE Trans. on Dielect. and Elect. Insul., Vol.8, No.4, 2001, pp. 652-657.5 G. Diaz, and A. Barbon, “Currents Sequence Analysis of a Transformer五分钟搞定5000字毕业论文外文翻译，你想要的
44、工具都在这里!在科研过程中阅读翻译外文文献是一个非常重要的环节，许多领域高水平的文献都是外文文献，借鉴一些外文文献翻译的经验是非常必要的。由于特殊原因我翻译外文文献的机会比较多，慢慢地就发现了外文文献翻译过程中的三大利器：Google“翻译”频道、金山词霸（完整版本）和CNKI“翻译助手。具体操作过程如下： 1.先打开金山词霸自动取词功能，然后阅读文献； 2.遇到无法理解的长句时，可以交给Google处理，处理后的结果猛一看，不堪入目，可是经过大脑的再处理后句子的意思基本就明了了； 3.如果通过Google仍然无法理解，感觉就是不同，那肯定是对其中某个“常用单词”理解有误，因为某些单词看似很简
45、单，但是在文献中有特殊的意思，这时就可以通过CNKI的“翻译助手”来查询相关单词的意思，由于CNKI的单词意思都是来源与大量的文献，所以它的吻合率很高。 另外，在翻译过程中最好以“段落”或者“长句”作为翻译的基本单位，这样才不会造成“只见树木，不见森林”的误导。四大工具： 1、Google翻译： google，众所周知，谷歌里面的英文文献和资料还算是比较详实的。我利用它是这样的。一方面可以用它查询英文论文，当然这方面的帖子很多，大家可以搜索，在此不赘述。回到我自己说的翻译上来。下面给大家举个例子来说明如何用吧比如说“电磁感应透明效应”这个词汇你不知道他怎么翻译，首先你可以在CNKI里查中文的，根据它们的关键词中英文对照来做，一般比较准确。 在此主要是说在google里怎么知道这个翻译意思。大家应该都有词典吧，按中国人的办法，把