1、 A Wavelet Based Approach for Fast Detection of Internal Fault in Power Transformers The power transformer is one of the most expensive elements of power system and its protection is an essential part of the overall system protection strategy. The differential protection provides the best protection
2、 for power transformer. Its operation principle is based on this point that the differential current during an internal fault is higher than normal condition. But, a large transient current (inrush current) can cause mal-operation of differential relays. Then, studies for the improvement of the tran
3、sformer protection have focused on discrimination between internal short circuit faults and inrush currents in transformers. The magnetizing inrush current has a large second order harmonic component in comparison to internal faults. Therefore , some transformer protection systems are designed to ha
4、lt operating during the inrush current by sensing this large second order harmonic. The second harmonic component in the magnetizing inrush currents tend to be relatively small in modern large power transformers because of improvements in the power transformer core materials. Also , it has been seen
5、 that the fault current can contain higher second order harmonics than the inrush current due to nonlinear fault resistance, CT saturation .the distributed capacitance in the transmission line, which transformer is connected to, or due to the use of extra high voltage underground cables. Various met
6、hods have been suggested for overcoming this protection system mal-operation. This paper presents a wavelet based method for discrimination among inrush current, internal short circuit ,external short circuit and energizing and it is not affected by CT saturation and it is able to detect internal fa
7、ults while transformer energization. Unlike Artificial Neural Network and Fuzzy logic based algorithms. This approach is not system dependent. The operating time of the scheme is less than 10ms. The Daubechies mother wavelet is used with a sample rate of 5 kHz. Then , the differential currents of th
8、e three phases are decomposed into two details and only the second level will be considered by using db5 mother wavelet. Discrete Wavelet Transform The wavelet transform is a powerful tool to extract information from the non-stationary signals simultaneously in both time and frequency domains. The a
9、bility of the wavelet transform to focus on short time intervals for high-frequency components and long intervals for low-frequency components improves the analysis of transient phenomena signals. Various wavelet functions ,such as Symlet,Morlert and Daubechies are used to analyze different power sy
10、stem phenomena. The mother wavelet must be selected performed based on its application and the features of signal .which should be processed. In this paper, Daubechies wavelet is used. There are three types of wavelet transform. Which are Continuous Wavelet Transform(CWT). Discrete Wavelet Transform
11、(DWT) and Wavelet Packet Transform(WPT). DWT is derived from CWT. Assume that x(t) is a tome variable signal, then the CWT is determined by (1): dtttxC W T )()(),(21 (1) Where, and are translating and scaling parameters, respectively. Also , )(t is the wavelet function and )(t is the complex conjuga
12、te of )(t . Wavelet function must satisfy(2) and should have limited energy: 0)( dtt (2) Then ,the discretized mother wavelet is as follows : )(1)(0000, mmmnmnbtt (3) Where, 0a 1 and 0b 0 and they are fixed real values. Also ,m and n are positive integers. DWT is expressed by (4): )()(),(, kkfnmfD W
13、 T k nm (4) Where, )(, knm is the complex conjugate of )(, knm. In (4), the mother wavelet is dilated and translated discretely by selecting and b . m0 and mnbb 00 (5) DWT can be easily and quickly implemented by complementary low pass and high-pass filters. Proposed Algorithm In the proposed algori
14、thm, the DWT is applied to the differential currents of three phases. The Daubechies Db-5 type wavelet is used as the mother wavelet and the signals are decomposed up to the second-level. Then , the spectral energy and standard deviation of the decomposed signals in the nd2 level are calculated. The
15、 proposed method consists of two steps； detection and discrimination. Disturbance Detection Under normal conditions and external faults, the differential currents have smaller values than internal faults. However in some operating conditions, the external faults can result in high differential curre
16、nts due to ratio mismatch of CTs or tap changes of power transformer. Then ,these conditions may cause mal-operation of the relay. Therefore ,a threshold current is used in order to prevent malfunctions caused by non-faulty currents. If one of differential currents exceeds this threshold value, it w
17、ill be identified as a fault. The threshold value id defined, as follows: 2 )( se c.d e t CTperCT iiki (6) Where CTi sec and CTperi are the secondary and primary CT currents, respectively, and k is the slope of the differential relay characteristic. If difii det, then the detection algorithm defines
18、 it as an internal fault. Disturbance Discrimination In order to classify disturbances, the differential currents are decomposed up to the second level, using Daubechies Db5 type wavelet with data window less than the half of the power frequency cycle. A sampling rate of 5 kHz, is considered for the
19、 algorithm(i.e . 100 samples per power frequency cycle based on 50 Hz). Then, the energy and standard deviation in the second detail are calculated for each differential current. It is seen that the spectral energy as well as the standard deviation in nd2 level tends to have high values during inrus
20、h currents. Then , a discrimination index ( mdD )can be calculated by multiplying the spectral energy by standard deviation in the second detail for each differential current, as follows: ESTDDmd (7) Where, STD is the standard deviation in nd2 detail and E is its spectral energy. The STD can be determined using the following equation: MddS T DMnme a nn 122)(2 )( (8) Where, )(2nd is n-th coefficient from detail 2. meand2 is its mean value and M is
