Analysis of wind farm islanding experiment

Transcription

1 Downloaded fro orbit.dtu.dk on: Apr 4, 19 Analysis of wind far islanding experient Pedersen, Jørgen Kaas; Pedersen, Knud Ole Helgesen; Poulsen, Niels Kjølstad; Akke, Magnus Published in: I E E E Transactions on Energy Conversion Link to article, DOI: 1.119/ Publication date: Docuent Version Publisher's PDF, also known as Version of record Link back to DTU Orbit Citation (APA): Pedersen, J. K., Pedersen, K. O. H., Poulsen, N. K., & Akke, M. (). Analysis of wind far islanding experient. I E E E Transactions on Energy Conversion, 15(1), DOI: 1.119/ General rights Copyright and oral rights for the publications ade accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requireents associated with these rights. Users ay download and print one copy of any publication fro the public portal for the purpose of private study or research. You ay not further distribute the aterial or use it for any profit-aking activity or coercial gain You ay freely distribute the URL identifying the publication in the public portal If you believe that this docuent breaches copyright please contact us providing details, and we will reove access to the work iediately and investigate your clai.

2 11 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 15, NO. 1, MARCH Analysis of Wind Far Islanding Experient Jørgen Kaas Pedersen, Magnus Akke, Meber, IEEE, Niels Kjølstad Poulsen, and Knud Ole Helgesen Pedersen Abstract This paper deals with the probles related to an islanding experient perfored at Rejsby Hede in Denark. During the experient several interesting observations were ade in connection to distortion of voltages and currents. Observations were also ade in connection to variation of frequency and phase. In this paper the data are analyzed in three different anners and the results are related to the physics of the electric syste. Index Ters Deterination of distortion, frequency estiation, inductions achines, islanding, wind far. I. INTRODUCTION AS EARLY as 1918, C. P. Steinetz considered wind power as very interesting [1]. He foresaw the use of induction generators attached to the wind-wheel, feeding the power into a larger bulk syste with hydro and stea cogeneration. Today, after 8 yr, wind power has been successfully introduced in Denark. Further expansion plans will turn wind power into a ajor production source in Denark. At reote windy locations, the grid can be weak, wind ills are clustered together to wind fars. One special characteristic of wind power is the variability in power generation. The expansion plans call for tools to analyze voltage and frequency behavior, both at synchronized and islanding operation. Paper [] describes the dynaics and stability of both synchronous and induction wind turbine generators. As reported in [] there is also a risk for self-excitation at islanding operation of induction generators with capacitive copensation. Paper [4] is a survey of wind turbine integration into a weak grid. In this paper we use easureent data fro an experient a wind far was forced into an islanding situation. The test is part of a larger EU-project to test an Advanced Static Var Copensator (ASVC, see [5] or []). The easureent noise, haronics, phase jup in voltage at islanding ake the data useful to benchark frequency estiation algoriths. The islanding experient presented in this paper is interesting fro two perspectives. In the narrow perspective it is iportant to ensure that nearby custoers to this specific wind plant are not disturbed. In a broader perspective it is also iportant to understand the underlying dynaics behind the frequency and voltage behavior after disconnection. A relevant question is how to interpret the ter frequency at islanding. Do we iplicitly ean that frequency always refers to rotor speed, or shall we strictly define frequency Manuscript received June 1, 1998; revised October, J. K. Pedersen and K. O. H. Pedersen are with the Departent of Electric Power Engineering, Build. 5, Technical University of Denark, DK-8 Lyngby, Denark. M. Akke is with SYDKRAFT, Carl Gustavs Väg 1, S-59 Malö, Sweden. N. K. Poulsen is with the Departent of Matheatical Modelling, Build. 1, Technical University of Denark, DK-8 Lyngby, Denark. Publisher Ite Identifier S ()1-X. Fig. 1. Situation plan for Rejsby Hede wind far. as the derivative of the phase angle? At islanding the two interpretations give very different answers. One of the paper s purposes is to show that independent ethods give consistent results. This is iportant since the analysis is based on data easureents fro the real world and the true frequency is unknown. Typical use of frequency estiation in power systes is in protection schees against loss of synchronis, under- or overfrequency relaying, input signal for power syste stabilization and as a part of spectral estiation for periodic signals. The aterial in the paper is organized as follows. Section II provides background inforation on the wind far, the island field test and the easuring syste. Section III presents a preliinary analysis of the recorded data. Section IV shows how outliers rejection can be used to get a sooth frequency estiate that corresponds to the rotor speed. Section V uses a paraetric approach frequency and haronic content are estiated siultaneously. Here frequency is strictly interpretated as the rate of change in phase angle. II. BACKGROUND A. Description of the Windfar The islanding experient was carried out on a wind far in Denark at Rejsby Hede, located in the south of Jutland. The wind far, with 4 wind turbines, its electrical surroundings, and an ASVC is shown in Fig. 1. To understand what happens after islanding, we have to consider the balance of reactive power of the generators, the internal cabling and the external overhead line. Each generator ( kw/9 V/4 poles induction achine) has a no-load reactive power consuption of kvar, which partly is balanced by a copensating capacitor producing 15 kvar. A 9 V/15 kv transforer on each generator has a reactive power consuption of 1 kvar, which is negligible /$1. IEEE

3 PEDERSEN et al.: ANALYSIS OF WIND FARM ISLANDING EXPERIMENT 111 Fig.. Voltage and current in one phase. Fig.. Voltage and current in phases around islanding. Internal to the wind far, there are three different types of 15 kv cables used between the generators and the bus bars. In all 14.7 k of cables with a total capacitance of 4. uf/phase equal to 5 kvar. The. k overhead line fro the wind far site to the switch gear adds around 15 kvar. When the switch gear is tripped and the wind far is islanding the total deand for reactive power at 5 Hz aounts 1. MVar. In other words, the wind far is 8% copensated which should prevent self excitation fro taking place. These considerations have not taken the ASVC into account. During noral operation of the wind far the ASVC produces the lacking reactive power. But the ASVC was switched off at the oent of islanding and had no influence on what happened afterwards. B. The Measuring Syste Voltage and current in -phases were easured during the islanding experient. The easureents started a short tie before the connection to the grid was switched off and continued a few seconds p after this disconnection. The phase voltages around 15= or 8.7 kv were easured by special designed capacitive voltage dividers to assure satisfactory frequency characteristic. The easuring syste consists of a PC equipped with a khz, 1-bit, ultichannel A/D-converter and as front-end a signal conditioning circuit including an antialiasing filter. The sapling rate was set to 1.8 khz, or 5 easureents during a 5 Hz cycle. Fig. 4. RMS and frequency in one phase. III. PRELIMINARY ANALYSIS Fig. gives an overall view of the easureents in one phase of voltage and current. The islanding situation starts after approxiately.4 s. Iediately after that the current drops to one tenth of the original level and decays slowly to zero. The voltage aintains its level ore or less after islanding, but soe variations in aplitude and frequency are introduced. The voltage disappears 1 s later due to a forced, echanical braking of all the wind turbines. Fig. gives a closer view of the behavior around the oent of islanding. A phase shift is observed in the voltage when islanding starts and a clear haronic appears in the voltage after islanding. This is the fifth haronic which before islanding is.4%, but after islanding increases to.5% the fifth haronic in the current is 1.1% before islanding. The higher haronics Fig. 5. Angle of space vector of voltage deviation fro 5 Hz. seen in both voltage and current before islanding are caused by the ASVC which iediately stops operation when islanding occurs. Fig. 4 shows the variations in RMS-value and frequency in one phase of the voltage derived fro a detection of zero-crossings over the easureent period. The apparently abrupt change in frequency fro 5 5 Hz at the tie of islanding is created ainly by the phase shift. The ethod can not distinguish between a sudden jup in phase and frequency. An alternative way of looking at the behavior is displayed in Fig. 5. First, the voltage space vector (a coplex sinusoid or Cisoid) corresponding to the -phase voltages is calculated. At each sapling point the three phase quantities are converted to

4 11 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 15, NO. 1, MARCH -coponent by ultiplying the inputs by a atrix A, that is v (k) = A 4 v r(k) v s (k) 5 v (k) v t (k) A = r p 1 7 p 5 : By interpreting v as real part and v as iaginary part, we get the coplex sinusoid v(k) =v (k) +jv (k): (1) Next, the difference between the angle of this vector and the angle of a pure 5 Hz space vector is deterined, and this difference is plotted in Fig. 5. It is seen that the frequency of the voltage before islanding is exactly 5 Hz the noise is generated by higher haronics. At the tie of islanding a phase shift of approx. 1 takes place not in one go, but over ost of a 5 Hz cycle, which ay be caused by the phases not being disrupted at the sae tie. After that the curve oves away fro the 5 Hz case, which indicates a change in frequency. The oscillations are produced by the fifth haronic. IV. FREQUENCY ESTIMATION WITH OUTLIER REJECTION In this section, it is assued that power syste frequency is governed by generator speed. All generators have inertia and therefore frequency is postulated to be continuous. Often voltage is preferred as input signal to frequency estiation algoriths. Voltage is advantageous since the agnitude is often close to the noinal value. In contrast, current can vary over a uch broader range, and ore seriously, would not work at no-load situations. Despite this the use of voltage has soe drawbacks. For weak power systes, i.e., low short circuit power, sudden load changes show up as phase jups in the voltages. Many algoriths are sensitive to these jups and they often produce spikes in the frequency estiate. The spikes can be characterized as statistical outliers. We propose a filter that estiates the ean and standard deviation of the frequency estiate and use this to reject the outliers. The new filter is cobined with ethod [7] to give an algorith ore robust to phase jups. One way to reduce the large phase jups is to calculate the internal voltage in the generator and use this for frequency estiation, see [8]. However, this requires current easureents and knowledge of the ipedance towards the generator seen fro the easureent point. Here we consider the case only the voltage is easured and the voltage alone is used for frequency estiation. A three step procedure is described below. First the data is prefiltered in a FIR low pass filter with crossover frequency 4 Hz and window length 18. To get a sapling rate suitable for fundaental frequency estiation, the rate is deciated by a factor 8, giving N =saples per 5 Hz cycle. Fig.. Phase angles for two saples of y. The idea behind the frequency estiation ethod [7], and any others, is to use a coplex quantity and estiate the frequency fro phase angle changes. The coplex space vector, or coplex sinusoid, is the input to the deodulation part. A. Deodulation The coplex signal v(k) fro (1), is deodulated with a known coplex phasor z rotating at the noinal syste frequency f o in opposition to the input. The deodulated phasor y is calculated as, y(k) =v(k)z(k) with z(k) =cos k f + j sin k f f s f s f s is the sapling frequency and f o is the noinal syste frequency. B. Estiation of Phase Angle Change The phase shift of the coplex variable y is used to estiate the unknown frequency deviation. For each saple, y rotates an angle 1 as illustrated in Fig.. Fig. shows a direct way to calculate the phase angle change 1 per saple, that is 1 = arg(y k ) arg(y k1 ): However this can cause proble with large phase jups when the angles cross over 18 or borders. One siple way to avoid these proble is to calculate 1 fro 1 = arg(y k 1 y k1 ) is used for coplex conjugate. C. Postfiltering with Outlier Rejection Fig. 7 shows the estiate after being postfiltered in a third order Butterworth low-pass filter with a cut-off frequency at 15 Hz. Fig. 7 shows a large frequency spike at t = :4 s caused by the phase jup when the wind far is disconnected fro the external grid. This is not a frequency variation caused by changing generator speed. When frequency is interpretated as change in rotor speed this spike should be filtered away. One way is to use a linear filter with a large tie constant. However,

5 PEDERSEN et al.: ANALYSIS OF WIND FARM ISLANDING EXPERIMENT 11 Fig. 7. Frequency estiate after third order Butterworth LP filter. Fig. 9. The resulting frequency estiate after low-pass, outlier rejection and final low-pass filtering. As a final step, the estiate goes into a FIR low-pass filter designed in Matlab with f c =Hz and N =8. Fig. 9 shows the result in a frequency estiate. Fig. 8. Frequency estiate after outlier rejection filter (solid) and the rejection bounds (dash-dot). this slows down the algorith. An alternative approach has been tested the spikes are rejected by characterizing the as statistical outliers. The standard deviation of the frequency estiate is calculated for a sliding window with length N. The equation used is s 1 X f (k) = 1f N i 1fi 1 i=kn+1 1f i is the deviation fro the window ean, 1f i = f i f and the window ean value f is f = 1 N kx i=kn+1 One way to avoid extree values at abnoral conditions, such as zero input, is to liit f to be within a reasonable range, f in < f < f ax. Outliers are rejected by putting a bound on each new frequency estiate. It has to be within the interval f f, that is f i : f k ax[in[f k ; f + f ]; f f ]: The output fro the Butterworth filter shown in Fig. 7 has been filtered in this way. The window length was N = 9and the standard deviation f was liited between.1 and 5 Hz. The result is shown in Fig. 8 together with the bounds. Fig. 8 shows that the outlier rejection filter eliinates the large spikes at around.4 s. V. PARAMETRIC ESTIMATION Three different types of ethods have been applied in order to consolidate the interpretations. The last one presented here is based on a siple paraetric statistical ethod. The basic assuption in this contexts is that the signals involved are periodic and it is possible to Fourier expand the signals. More precisely it is assued that a odel y k =^y k + e k = r; s; t X n ^y k = a i cos(n i!k) +b i sin(n i!k) () i= is valid for each phase (yk is the voltage or current in one phase). Here k is the tie index and e k is a zero ean noise. The coefficients a i, b i contain inforation on the aplitude A i and the phase (with respect to origin) P i of each haronic coponent in each phase, i.e., ^y k = n i= A i sin(n i!k + i ). The basis assuption concerning the periodicallity of the signals are, according to Fig., clearly not fulfilled in the islanding point, which is not the sae for all three phases. The odel () ight also be written as x k = [ y r k y s k y t k ]=x> k +[er k e s k e t k ] () 4 cos(!k) sin(!k) cos(!k). sin(n n!k) 7 5 = 4 a r 1 a s 1 a t 1 b r 1 b s 1 b t 1 a r a s a t. b r n b s n b t n If the frequency! is known then it is a standard task to ake a estiate of the coefficients in () which iniize the su of squares, J 1 NX X j=1 7 5 : y k j ^y k j (4)

6 114 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 15, NO. 1, MARCH Fig. 1. The contents of higher haronic coponents in the voltage before and after islanding. Fig. 11. The aplitude of the first and fifth haronic coponent. over a data window of length N. Such an LS-estiate is siply given by ^ =(X > X ) 1 X > Y (5) X = 4 x> 1.. x > N y1 7 5 r y1 s y1 t Y = : yn r ys N yt N On the other hand, if the frequency! is unknown, it has to be estiated too. If we use the LS criterion (4) and define g = N X H l = X i=1 NX X ^y (y k ^y k ^y (y k ^y k ): [The derivatives can easily be found using ()] then the estiate of! and the paraeters in can found by utilizing a (variant of a) Newton Raphson ethod, each iteration obey! l+1 =! l g l () H l and (5). The approxiation involved when using (5) and () is to neglect the cross ter in the second derivative. There exists a recursive version of this ethod, which turns out to be a kalan filter echanization, and is well suited for real tie processing. In a post orte analysis (as used in this paper) the algorith is directly applicable with soe rectangular windowing techniques. According to Fig. the voltage have a high distortion coponent after islanding. In order to investigate this the paraeters in () are estiated up to the order of 1. The results are shown in Fig. 1, it is clear that the fifth haronic is rather significant after islanding. This is also clear fro Fig. 11 the variations of the aplitude of the first (i.e., base frequency) and fifth haronic in the voltage are shown. The estiate of the paraeters in () are based on data fro a tie window with a length corresponding to 1.5 of a cycle (at 5 Hz). Siilarly, before islanding there exist significant coponents in the currents, as shown in Fig. 1. At the point of islanding the frequency and phases change instantaneously due to the lack of electric oentu. Since Fig. 1. The contents of higher haronic coponents in the current before islanding. Fig. 1. The estiated frequency as function of tie. the wind turbine syste accelerates the frequency and phases vary. This is shown in Fig. 1, soe oscillation in the echanical syste is obvious. In order to investigate the instantaneous changes in frequency and phase the analysis is perfored just before and just after the islanding. The result of the analysis is an estiate of the basic frequency! b and! a (before and after) and phase b and a (with respect to origin) of the first haronic in the syetric syste. The analysis shows a jup in frequency of 1.15 Hz. The phase jup can be estiated by: =(! a t is + a ) (! b t is + b ) t is is the instant of islanding. The phase jup in the reported experient is 7.5.

7 PEDERSEN et al.: ANALYSIS OF WIND FARM ISLANDING EXPERIMENT 115 VI. DISCUSSION The above analyses focus on the behavior of the voltage during and after the oent of islanding. The reason for the behavior can be traced back to the electrical and echanical properties of the syste. A phase shift in the voltage ust be expected when islanding occurs. The size of the shift can be calculated when knowing the characteristic data of the generators and the ipedances in the grid and knowing that the generators were 75% loaded just before islanding. A phase shift of approxiately 1 is found which is in good agreeent with the easureents. Since the generators go fro an alost full-load to a no-load condition at the oent of islanding, then a shift in frequency takes place. However, this is just a inor shift due to the low slip of the generators. After islanding, the wind turbines start accelerating the generators which explains the increase in frequency observed in the easureents. The increase in frequency is superiposed with a slow oscillation which ight be caused by torsion in the shaft connecting two inertia loads on one side the wings of the turbine, and on the other side the rotor of the generator. It is surprising that the voltage does not fall away after islanding, since the generators are under-copensated. Iediately after islanding the voltage decreases, but after a short while it starts increasing again following the variations in the frequency. We have no sure explanation to this phenoena, but it ight be caused by the capacitance of the internal cabling and of the k long overhead line fro the wind far site to the line breaker. An over-copensation takes place if just 8 wind turbines or less are running due to this extra capacitance. Such a situation can arise if the brakes on all wind turbines are not activated at the sae tie. According to specifications breaking is initiated. s after a frequency higher than 51.5 Hz is detected. This sees not to be the case and this point has to be exained closer in future experients. The fifth haronic appearing clearly in the voltage after islanding is produced by the generators. Before islanding this fifth haronic coponent is suppressed because the wind far is connected to a strong low ipedance power syste. VII. CONCLUSIONS When planning a wind far, precautions should be taken against islanding to prevent disastrous consequences for coponents connected to the islanding network. The experient has deonstrated that the behavior around islanding is quite different fro what norally is seen in a power syste. Islanding starts with a sudden phase jup followed by changes in voltage level and frequency and significant haronics are appearing. New tools are necessary to analyze islanding. Different atheatical approaches have been tested in this paper and they all lead to consistent results. Thus, identifying islanding fro easureents of the -phase voltage is possible and suitable algoriths can be derived. The results have attracted attention aong Danish power utilities and wind turbines anufactors and we have been encouraged to continue the study. A obile easuring syste have been designed. New experients and further experiences will be gained in the future. REFERENCES [1] C. P. Steinetz, Aerica s energy supply, in Proc. of the AIEE., vol. XXV, 1918, pp [] E. N. Hinrichsen and P. J. Nolan, Dynaics and stability of wind turbine generators, IEEE Trans. on Power Apparaturs and Systes, vol. PAS-11, no. 8, pp. 4 48, Aug [] G. L. Park and O. W. Zastrow, Interconnection issues concerning consuer-owned wind electric generators, IEEE Trans. on Power Apparaturs and Systes, vol. PAS-11, no. 7, pp. 75 8, July 198. [4] T. S. Davies, R. Mayer, and H. Nouri, Integration of wind turbines into the low voltage networks, in th Universities Power Engineering Conference, vol., 1995, pp [5] J. Kaas Pedersen, Copensating of reactive power fro wind turbines with power electronics equipent, in EWEA Special Topic Conference on Integration of Wind Power Plants in the Environent and Electrical Systes, Ro, 199. [] K. Bergann, K. Renz, F. Schettler, R. Stöber, H. Tyll, M. Weinhold, K. Søbrink, N. Jenkins, and J. K. Pedersen, Application of GTO-based SVCS for iproved use of the Rejsby Hede windfar, in Ninth National Power Syste Conference, Kandur, 199. [7] M. Akke, Frequency estiation by deodulation of two coplex signals, IEEE Trans. on Power Delivery, vol. 1, no. 1, pp , Jan [8]. Sauelsson, Power syste daping Structural aspects of controlling active power, Ph.D. dissertation, Departent of Industrial Electrical Engineering and Autoation, Lund Institute of Technology, Lund, Sweden, [9] C. P. Steinetz, Aerica s energy supply, Proc. of the IEEE, vol. 8, no. 4, pp. 7 74, Apr Jørgen Kaas Pedersen was born in a village near Copenhagen, Denark, on March, 195. He graduated with a Masters degree in applied physics in 191. After working eight years in Danish industries, he was eployed at the Technical University in a departent for electric technique. His interest for power electronics increased, especially for use of power electronics in the grid. He is now eployed in the Departent of Electric Power Engineering and he is working with ASVC for copensating reactive power in the grid and probles in connection with large wind fars connected to the Danish grid. Magnus Akke (M 9) was born in Lund, Sweden, in 191. He received the M.E.E., the Licentiate, and the Ph.D. degrees fro Lund Institute of Technology, Sweden, in 198, 1989 and 1997, respectively. He also has a Bachelors degree in business adinistration fro Lund University, Sweden. In 199, he started at Sydkraft, a Swedish power utility his work tasks are power syste analysis and relay protection. He has been a visiting Scientist at University of Newcastle, Australia, and at Cornell University, USA. Niels Kjølstad Poulsen was born in the central part of Sjaelland, Denark in 195. He received his M.Sc. and Ph.D. degrees in electrical engineering fro the Institute of Matheatical Statistics and Operations Research (IMSOR), at the Technical University of Denark, in 1981 and 1984, respectively. He has been eployed at the Technical University of Denark since 1984, and since 199 he has been an Associate Professor at the Departent of Matheatical Modelling. His priary research interests are within stochastic control theory, syste identification, recursive estiation, and adaptive control. Knud Ole Helgesen Pedersen was born in Copenhagen, Denark in 194. He received the M.Sc. degree in electrical engineering in 197 and the Ph.D. degree in autoatic control in 197, both fro the Technical University of Denark. Fro 197 to 197 he worked at CERN, Geneva, and fro 197 to 1985, he worked in Danish industry as a Research Engineer. During that tie he was ainly occupied with digital process control. Since then, he has been an Associate Professor at the Technical University of Denark with his ain interest in icrocoputer applications in the power syste.