过电压保护

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 2, APRIL 2008 657 Overvoltage Protection of Large Power Transformers—A Real-Life Study Case Claus Leth Bak, Kristin Erla Einarsdóttir, Einar Andresson, Jesper M. Rasmussen, Jan Lykkegaard, and Wojciech Wiechowski Abstract—This paper demonstrates the results from a detailed study of the overvoltage protection of a particular 400/150-kV 400-MVA power transformer. The work presented here is based on a real-life power system substation design and data and initiated by Danish TSO Energinet.dk as a consequence of serious trans- former overvoltage damage. A simulation model for the entire system consisting of overhead line, transformer, surge arrester, and earth grid has been created in PSCAD/EMTDC. The main focus has been put on the earth grid, which has been submodeled in detail in MATLAB using an electromagnetic transient approach based on the thin-wire program made by J. H. Richmond for NASA in 1974. The earth grid model is verified with excellent agreement compared to already published results. The overvoltage performance of the particular case is analyzed, and it shows that the transformers LIWL have probably been exceeded. It is clearly illustrated that the transient performance of the earth grid plays an important role in the overall overvoltage protection system design. Index Terms—Dynamic resistance, earth grid design, light- ning impulse withstand level (LIWL), MATLAB, overvoltage protection, overvoltage protection simulation, PSCAD/EMTDC, transient behavior. I. INTRODUCTION ON June 18, 2002, a heavy thunderstorm swept over NorthJutland, Denmark, resulting in a serious fault in Energinet. dk’s 400/150-kV transformer placed at the Nordjyllandsværket 400-kV transformer station (NVV5). According to Energinet.dk, the fault was caused by a light- ning transient on the 150-kV transmission grid. Apparently, the transient lightning voltage exceeded the LIWL of the transformer. Fig. 1 illustrates the record breaking amount of lightning over Denmark on June 18, 2002. Manuscript received September 6, 2006; revised February 5, 2007. Paper no. TPWRD-00522-2006. C. L. Bak is with the Institute of Energy Technology (IET), Aalborg Univer- sity, Aalborg East DK-9220, Denmark (e-mail: clb@iet.aau.dk). K. E. Einarsdóttir is with Rafhönnun, Reykjavik 108, Iceland.. E. Andresson is with RTS Electrical Engineering Consultants, Reydarfjordur 730, Iceland. J. M. Rasmussen is with Aalborg Kommune Elforsyningen, AKE, Aalborg 9000, Denmark. J. Lykkegaard and W. Wiechowski are with Energinet.dk (TSO), Frederica 7000, Denmark. Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TPWRD.2007.905793 Fig. 1. Intensity of lightnings over Denmark on June 18, 2002. This incident has caused speculations within Energinet.dk about the effectiveness of the lightning protection of the trans- formers now used at Energinet.dk’s power stations. The possi- bility of this occurring again to any of the other power trans- formers in Eltra’s possession is a major concern. The main con- cern of the project is to make a simulation model of that part of the substation which surrounds the transformer (see Fig. 2), and to simulate a double exponential lightning impulse current directly on a phase line, which will propagate toward the trans- former in the form of a travelling wave. The main emphasis will be put on investigating the overvoltage distribution in the system with respect to the LIWL of the transformer and to simulate the components that are most likely to have caused excessive LIWL, thereby damaging the transformer. These are the 150-kV surge arresters, the earth grid with respect to GPR, and the transformer itself. The 150-kV overhead line between the 150-kV substation NVV3 and the 400-kV substation NVV5 is included in the sim- ulation. The results will then be used to determine a possible weakness in the overall overvoltage protection design. This paper presents a description of the real-life power system with sufficient details to study the overvoltage protection of the power transformer in detail and a description of the trans- former damage. A number of causes capable of resulting in such damage is listed and a hypothesis is postulated. Hypothesis: The transformer was not adequately protected at the 150-kV side, so the lightning insulation withstand level (LIWL) was exceeded. The action of this hypothesis was to model the system (Fig. 2) in such detail that a realistic simulation of the overvoltage pro- tection behavior could be performed and, in this way, spread 0885-8977/$25.00 © 2007 IEEE 658 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 2, APRIL 2008 Fig. 2. Overview of the system, with the overhead line, the surge arrester, the transformer, and the earth grid. Fig. 3. Photograph of the 400/150-kV power transformer and its nearest sur- roundings at the 400-kV substation NVV5. some light on the possible cause of transformer damage. The simulation model is further used to analyze possible improve- ments of the overvoltage protection, mainly regarding the de- sign of the earth grid. This will be presented in another paper. II. SYSTEM DESCRIPTION A. Substation The damaged power transformer is located at a normal out- door switchyard with proper shielding of both 150- and 400-kV overhead line connections plus grounding systems and rods at the entire substation area. Fig. 3 shows a photograph of the trans- former location at the substation. Fig. 4 shows the configuration in a scalable drawing so the connection of surge arresters and shielding can be identified: 1) LV surge arrester, 2) HV surge arrester, 3) transformer, 4) coolers, 5) suspension tower, 6) insulator, 7) ground wires, 8) HV phase conductor, and 9) LV phase conductor. Further details can be found in [1] which is a Masters Thesis elaborated by K. E. Einarsdottir, E. Andresson, and J. M. Rasmussen. This paper presents the main results of parts of their work. B. 400/150-kV Transformer The transformer is a three-phase 400-MVA ASEA oil-im- mersed autotransformer with data according to Table I. Bushings have a higher LIWL than transformers (LV side 750 kV and HV side 1675 kV). The transformer age is 14 years. Fig. 4. Configuration of transformer installation. TABLE I 400/150-kV ASEA AUTOTRANSFORMER DATA TABLE II 150-kV SURGE ARRESTER DATA Prehistory: Inservice always and very lightly loaded. Oil and gas analysis has previously shown no sign of premature aging. C. Surge Arresters Only the 150-kV surge arrester data are listed (as the over- voltage is assumed to originate from the LV side). These are ASEA XAR 170-A3 with data according to Table II and a pro- tective characteristic according to Fig. 5. D. Earth Grid The surge arresters are connected to the same earth system as the power transformer, although no direct connection be- tween surge arrester ground terminal and transformer exists, which is recommended in certain literature (i.e., [2] and trans- former manufacturer ABB). The earth grid is a slightly irreg- ular meshed grid of approximately 140 135 m or about 19 000 m in size. It is made of 95 m bare stranded copper wires and is buried at a depth of approximately 1 m. Earth rods are BAK et al.: OVERVOLTAGE PROTECTION OF LARGE POWER TRANSFORMERS 659 Fig. 5. ASEA XAR 170-A3 surge arrester data. Fig. 6. Earth grid in the surrounding of the transformer. The 150-kV connection is toward the top of the figure. Figure scalable with coordinates shown to the left. Figure shows only part of the earth grid. located at the periphery of the earth grid at an 8–42-m distance from each other. These are of the type Elpress, each consisting of a 6-m-long steel pipe and a 95-mm copper wire and located at a minimum of 1 m from each foundation as required by the IEC-1024-1 standard [3]. Fig. 6 shows that the transformer and the surge arresters are positioned at the outskirts of the earth grid, with the 150-kV surge arresters located only 8 m from the periphery. The squares and irregular boxes are the equipment foundation blocks. All three surge arresters are interconnected forming a relatively large mesh size to the periphery of 21 7.5 m and a mesh size of 4 16 m toward the transformer. The surge arresters are connected to the transformer neutral point as may be seen in Fig. 6, with a conductor length of 15 m from the phase A surge arrester, grid depth included. No earth rods are located very close to the surge arresters. The transformer is mounted on support units over a well that will drain any oil spill from the transformer. According to En- erginet.dk, gravel may have been used as a fill up material when mounting the transformer and the surge arresters. This gravel may therefore embrace the earth conductors between the trans- former and the surge arresters. The dynamic behavior of the earth system with respect to lightning impulses is the main focus of this project and is described in Section III. The static resis- tance of the entire earth grid is calculated based on the Schwartz Fig. 7. Damaged transformer. (a) The three phases with the faulty phase fur- thest to the left. (b) The faulty winding seen from the outside after various paper layers have been removed. Fig. 8. (a) Electrical diagram showing where the fault has occurred and (b) the fault occurring between two layers in the series winding. equation from IEEE-80 [4] and amounts to with specific resistivity of the soil . E. Transformer Damage The lightning activity in Denmark on June 18, 2002 was very heavy. There were more than 110 000 lightnings over Denmark that day and over 10 000 lightnings in an area with a radius of 50 km around the transformer substation. About 8500 positive and negative sky-to-earth lightnings were registered and 4%–5% of these had an amplitude of more than 30 kA and nearly all were negative (99%). Some of these were located (taking the accuracy of lightning detection systems into consideration) very close to the overhead lines of the substation. Fig. 7. shows the damage after opening the transformer at the ABB factory. The trans- former winding connection (autotransformer) is shown in Fig. 8, where it is seen that the fault occurred between two layers of the series winding. After disassembling, the transformer was re- paired and put back into service after approximately one year. III. MODELLING AND SIMULATION In order to simulate the overvoltage amplitude at the trans- former terminals, models are created from a transformer, surge arrester, earth grid, and overhead line. These are combined in a model of the total system implemented in the PSCAD/EMTDC software together with a double-exponential lightning surge source. The models of each component will be discussed briefly in the following sections. Further explanations to the models, especially concerning the earth grid model, can be found in [1] and is intended to be the main topic of a future paper. A. ASEA Autotransformer Model The transformer must be modeled sufficiently to possess terminal properties, which reflects its high-frequency behavior 660 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 2, APRIL 2008 Fig. 9. Transformer winding with one end grounded. (a) l is the total length of the winding and x is the distance from the top of the winding to an arbitrary point in the winding. (b) Equivalent electrical circuit of the winding [6]. Fig. 10. Capacitance values for the autotransformer windings (ABB). sufficiently to achieve realistic results of overvoltage stresses. Transformers are normally modeled [5] as a single capacitance from line terminal to ground. More detailed models are nor- mally used for studying the internal voltage distribution of the windings. This work uses an approach originally proposed by [6], which represents each phase winding as one single winding possessing capacitive, inductive, and resistive behavior as illustrated in Fig. 9. Depending on available transformer construction data and the need for a very precise model, the concept in Fig. 9 can be less or more complex (i.e., the degree of “lumpedness” and the in- clusion of self- and mutual (between parts of the winding) in- ductances and resistive damping). This work has used a combi- nation of capacitances (originating from ABB data, see Fig. 10) and resistances and inductances calculated on a simplified rep- resentation of the transformer geometry. The main purpose of extending the transformer model to in- clude both inductance and resistance was chosen in order to verify the high-frequency terminal behavior of the transformer, as this is very important for reliable results in the complete model. This was accomplished by implementing the model [1] in PSCAD and simulating the same situation as the transformer is tested against it in the factory test. This is the Hagenguth test Fig. 11. Model proposed by Fernandez. Fig. 12. Surge arrester model and double exponential test circuit in PSCAD. [7, p. 165], which impresses reduced, full, and chopped light- ning impulse voltages to the transformer terminals and measures the ground return current to check for damages that occurred during the testing. The factory test was available for the present transformer and this approach was carried out with satisfactory results [1], which supported the belief of a sufficient model of the transformer, although it is quite complicated to obtain results with more than just the main features (rise time, peak value, and decaying) of the ground return current close to the actual test results. This model consists of 63 partial lumped capacitances, 26 lumped inductances (air core assumed concerning high-fre- quency behavior), and 26 resistances for each phase. B. ZnO Surge Arrester The nonlinear surge arrester dynamics are modeled using the approach proposed by [8], which is a simplified model of the IEEE model with model parameters described as proposed in [9] and [10]. Fig. 11 shows the Fernandez approach. represents the inductance in the electric path through the ZnO blocks and is determined using the dimensions of the surge arrester. and represent the nonlinear resistivity of the ZnO blocks and can be estimated from the surge arrester residual voltage, see Section II-C. represents the terminal capacitance of the surge arrester. R is included to avoid numerical instability. The surge arrester model (see Fig. 12) is verified against manufacturer residual voltage data and excellent agreement was achieved (maximum 1.5% error). BAK et al.: OVERVOLTAGE PROTECTION OF LARGE POWER TRANSFORMERS 661 C. Earth Grid The purpose of making a model of the earth system is to cal- culate the voltage between the surge arrester ground terminal and the neutral point of the transformer, which results from a difference in ground potential rise (GPR) under the two com- ponents, when a lightning current surges through the surge ar- rester into the earth grid. An electromagnetic-field approach is the best choice when the need for calculation of transient volt- ages between points of the earth grid is present [11]. The earth grid model is a transient electromagnetic program written in the C-based programming language of MATLAB. It is based on the thin-wire structure program originally written in Fortran code by J. H. Richmond [12], [13]. The model performs an electro- magnetic analysis on wire structures in the complex frequency domain, based on closed-form expressions and Simpson’s rule of integration for the solution of electromagnetic fields. Its func- tion is to determine the electric near fields at the surface of the wire structure, due to the longitudinal current flowing in each section of the wire. The electric-field calculation is then used to determine the dynamic impedance, both self and mutual, of the wire structure in order to determine the current distribution in the overall grid. The grid is divided into segments and the current distribution is approximated by defining every two seg- ments as a dipole with a piecewise sinusoidal current distribu- tion given with sinusoidal expansion functions, as it is very close to the natural current distribution on a perfectly conducting thin wire. A sinusoidal dipole is used as a test source, as this is prob- ably the only finite-line source with simple closed-form expres- sions for the near-zone fields, and the mutual impedances be- tween two sinusoidal dipoles may be determined from exponen- tial integrals [13, p. 7]. The thin-wire approach has been used by Grcev [11], [14]–[17] to determine the electric fields in earth grids caused by lightning surge currents. Grcev refers to Rich- mond’s thin wire program in [15, p. 394], but he additionally in- cludes image theory in his model to account for reflections due to interface of air and earth, as this is not included in Richmond’s program. Grcev also describes in his articles how to implement an injected current, also not included in Richmond’s program. As Richmond’s thin-wire program was not specifically designed for calculating electromagnetic fields in earth grids, the program needed to be adapted to the problem presented in this report. All unnecessary functions to the presented problem have been eliminated from the program, which now has the main function of calculating antenna problems in a homogeneous conducting medium. Reflections of the electric field due to the interface of air and earth have been taken into consideration with the mod- ified image theory, and to make injection of surge current pos- sible, the modifications suggested by Grcev have been imple- mented in the program. Only the front time of the current wave is of interest since this provides the highest frequency and thereby the highest electric fields. All simulations are therefore made in the frequency domain, using the frequency corresponding to the desired current front time at each time, and a conversion of the current wave from the time domain to the frequency do- main by Fourier transforms is therefore not needed. The basic model (before implementing modified image theory and injec- tion current) has been verified thoroughly with results presented Fig. 13. Linear electrode energized at one of the ends. in Richmond’s notes [12]. After implementation of the modified image theory and the injection current, the model was verified by comparing the results with the results presented in [15] with very good agreement. The following assumptions and limita- tions are made in the model of the earth grid. 1) The wire structure is made of straight cylindrical metallic conductors. 2) The wire is subject to the thin-wire approximation, and the conductor radius is therefore assumed to be much smaller than the wavelength, with the wire length being much greater than the wire radius (At least 30 times greater [12, p. 12]). 3) Image theory is applied to compensate for the effects of a ground plane (i.e., the interface between air and earth is taken into consideration). This limits the frequency range of the model to a few megahertz [16]. 4) The media of earth and air are assumed to be homogeneous with a horizontal ground plane boundary between them. 5) The current on wire ends is assumed to be zero. 6) For accuracy, the longest wire segment should not greatly exceed 1/4 wavelength [13]. 7) Soil ionization is not taken into consideration. The MATLAB-made program is called TEMP and details can be found in [1] and a future paper. Verification is performed against two situations: 1) 15-m-long Horizontal Electrode: According to [15], ver- ification is carried out against a 15-m-long horizontal electrode (Fig. 13) buried at 1-m depth (soil resistivity 2000 m) with a wire radius of 0.007 m with energization by injecting time-har- monic currents of 1 A at three different frequencies: 50 Hz, 2.247 MHz, and 6.741 MHz. Results from TEMP are compare