AckermannTh. (ed) Wind Power in Power Systems
Подождите немного. Документ загружается.
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 17_CHA16 .3D – 361 – [349–364/16]
17.12.2004 10:38PM
substantially affect the voltage of the low-voltage grid. The voltage transient can disturb
sensitive equipment connected to the same part of the grid.
The amplitude of the current originating from switching in an unloaded capacitor is
determined by the impedance of the grid and the capacitance of the capacitor. The
frequency, f, of the transient can be determined approximately by:
f ¼
1
2
1
LC
; ð16:3Þ
where L is the indu ctance of the grid and C is the capacitance of the capacitor.
In order to improve the calculations of the connecting current and voltage, a more
detailed model must be used. The Electro Magnetic Transie nt Pr ogram (EMTP) makes
it possible to use frequency-dependent parame ters. Calculations of switching transients
on a low-voltage grid equipped with two wind turbines are presented in Larsson and
Thiringer (1995).
16.6 Frequency
On the one hand, the incorporation of a relatively small amount of wind power into the
utility grid normally does not cause any interfacing or operational problems. The
intermittent power production from wind turbines is balanced by other production
units (Hunter and Elliot, 1994). On the other hand, the effect of wind power is very
important in autonomous power systems (see also Chapter 14). In an autonomous grid,
the spinning reserve is supplied by diesel engines, and it is usually small. This small
spinning reserve will give rise to frequency fluctuations in case of a sudden wind rise or
wind drop. Hence, in a wind–diesel system, voltage and frequency fluctuations will be
considerably larger than in an ordinary utility grid.
Over the past decade, different types of wind turbines and wind–diesel systems for
autonomous grids have been tested. Type A wind turbines used to be the most common.
49.0
49.5
50.0
50.5
51
18.19
20.09
21.59
23.49
01.39
03.29
05.19
Time
Frequency, f (Hz)
Figure 16.15 Frequency variations during two nights. During one night the wind turbines (WTs)
operated (gray line) and during the other night they were shut off owing to low winds (black line)
Wind Power in Power Systems 361
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 17_CHA16 .3D – 362 – [349–364/16]
17.12.2004 10:38PM
Today, an increasing number of D1 types are used. Figure 16.15 shows measurements at
a wind–diesel system with a relatively small amount of wind power, during two nights.
The installed wind power corresponded to approximately 10 % of the total diesel
power installed on the island. The frequency on the grid was measured during two
nights, one night with wind turbines and one night without. There were two frequency
drops during the night when the wind turbines were not operating (black line). These
two drops were most likely caused by the stop of one or several diesel engines. The other
curve (gray line), which represents the frequency with operating wind turbines, shows an
increase in frequency. The frequency exceeded 50 Hz throughout the night, which
indicates that some diesel engines ran at low load. A likely explanation is that the utility
company was afraid to stop too many diesel engines in case the wind suddenly dropped.
If the portion of wind power is to be increased further (i.e. if the wind–diesel system is to
operate solely on wind power under high-wind conditions) the power from the wind
turbines must be controllable. Figure 16.16 shows measurements from such a specially
designed wind–diesel system, which uses a Type D1 wind turbine.
The figure shows the power from the wind turbine and the frequency measured during
one night. The figure illustrates that the turbine was switched off and did not produce
any power during the first 1.5 h. During this time, the plant operated in diesel mode. The
plant then worked in a mixed mode and the wind turbine operated together with the
diesel for approximately 4 h. After this, and for the rest of the night, the wind speed was
high enough for the wind turbine to operate alone. The total power consumption was
relatively constant, slightly exceeding 15 kW. The criterion for operating the plant in
wind mode is that the rotational speed of the wind turbine exceeds a predetermined
value, in this case 60 rpm.
The frequency rose from approximately 48 Hz in the diesel mode to 50 Hz in the
mixed and wi nd modes. The diesel seems to have a governor with a frequency of 52 Hz
at no load and up to 48 Hz at full load. For the rest of the night, the plant ran in the
46
47
48
49
50
51
0
5
10
15
20
25
21.11
23.01
00.51
02.41
04.31
Time
Frequency, f (Hz)
Power, p (kw)
Figure 16.16 Frequency variations (black line) and power output from a Type D1 wind turbine
(gray line) during one night
362 Practical Experience with Power Quality
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 17_CHA16 .3D – 363 – [349–364/16]
17.12.2004 10:38PM
wind mode. As can be seen, the frequency was very stable during that time. In fact, the
frequency was much more stable in the wind mode than in the other two modes.
16.7 Conclusions
This chapter investigated the power quality of grid-connected wind turbines. Further-
more, it focused on the characteristics of fixed-speed and variable-speed wind turbines.
From an electrical point of view, wind turbines may be divided into tw o main groups –
fixed-speed an d variable-speed. Both groups of wind turbines have advantages and
disadvantages regarding their interaction with the grid and power quality. Wind tur-
bines have an uneven power production following the natural variations in the wind.
Uneven power production is the same for all kinds of wind turbines. Each time a turbine
blade passes the tower, it enters into the tower shadow. If the turbine operates at fixed
speed, the tower shadow and wind speed gradients will result in fluctuating power. Both
uneven power production and power fluctuations cause voltage variations. Load-flow
calculations can be used to calculate slow variations in the voltage caused by the uneven
power production of wind turbines. The power fluctuations of the wind turbine may
cause flicker disturbances. In order to calculate the impact of flicker, measurements and
subsequent flicker calculations must be performed.
Apart from possible oscillations between the grid impeda nce and the shunt capacitor
banks for power factor correction, fixed-speed win d turbines do not produce any
harmonics. When it comes to variable-speed wind turbines, however, the situation is
different. Depending on the type of inverter used, different orders of harmonics are
produced.
Transients seem to occur mainly when wind turbines are started and stopped. A large
in-rush current and thereby a voltage dip can be avoided if the wind turbi ne is equipped
with a soft-starter. When the shunt capacitor bank is switched in, a large current peak
occurs. The current peak may substantially affect the voltage on the low-voltage side of
the transformer.
In an autonomous grid supplied by diesel engines, the spinning reserve is limited and
gives rise to frequency fluctuations when fast load changes oc cur. Hence, the frequency
of an autonomous grid is normally not as stable as that of a large grid. When wind
power is incorporated into an autonomous grid, a sudden rise or drop in wind wi ll affect
the power balance, with frequency variations as a result. The use of sophisticated
variable-speed wind turbines can eliminate this problem and actually improve the
frequency balance.
References
[1] Ballard, L. J., Swansborough, R. H. (Eds) (1984) Recommended Practices for Wind Turbine Testing: 7.
Quality of Power. Single Grid-connected WECS, 1st edn, Elsgaards bogtrykkeri.
[2] CENELEC (European Committee for Electrotechnical Standardisation) (1994) Voltage Characteristics of
Electricity Supplied by Public Distribution Systems, EN 50 160, CENELEC.
[3] Demoulias, C. S., Dokopoulos, P. (1996) ‘Electrical Transients of WT in a Small Power Grid’, IEEE
Transactions on Energy Conversion 11(3) 636–642.
Wind Power in Power Systems 363
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 17_CHA16 .3D – 364 – [349–364/16]
17.12.2004 10:38PM
[4] Gardner, P. (1993) Flicker from Wind Farms, in Proceedings of the BWEA/SERC RAL Workshop on
Wind Energy Penetration into Weak Electricity Network, Rutherford, UK, June, pp. 27–37.
[5] Hunter, R., Elliot, G. (1994) Wind–Diesel Systems, Cambridge University Press, Cambridge, UK.
[6] IEC (International Electrotechnical Commission) (1990a) ‘Flickermeter – Functional and Design Speci-
fications’, IEC 60868, IEC.
[7] IEC (International Electrotechnical Commission) (1990b) ‘Amendment 1 to Publication 60868,
Flickermeter – Functional and Design Specifications’, IEC.
[8] IEC (International Electrotechnical Commission) (1991) ‘Electromagnetic Compatibility, General Guide
on Harmonics and Inter-harmonics Measurements and Instrumentation’, IEC 61000-4-7, IEC.
[9] IEC (International Electrotechnical Commission) (1997) ‘Amendment 1 to Publication 61000-4-7, Elec-
tromagnetic Compatibility, General Guide on Harmonics and Inter-harmonics Measurements and
Instrumentation’, IEC.
[10] IEC (International Electrotechnical Committee) (1998) ‘Power Quality Requirements for Grid Connected
WT’, IEC 61400-21, Committee Draft, IEC.
[11] Larsson, A. (2002a) ‘Flicker Emission of Wind Turbines During Continuous Operation’, IEEE Transac-
tions on Energy Conversion 17(1) 114–118.
[12] Larsson, A. (2002b) ‘Flicker Emission of Wind Turbines Caused by Switching Operations’, IEEE
Transactions on Energy Conversion 17(1) 119–123.
[13] Larsson, A
˚
., Thiringer, T. (1995) ‘Measurements on and Modelling of Capacitor-connecting Transients
on a Low-voltage Grid Equipped with Two WT’, in Proceedings of the International Conference on Power
System Transients (IPST ‘95), Lisbon, Portugal, September, pp. 184–188.
[14] Reid, W. E. (1996) ‘Power Quality Issues – Standards and Guidelines’, IEEE Transactions on Industry
Applications 32(3) 625–632.
[15] Saad-Saoud, Z., Jenkins, N. (1999) ‘Models for Predicting Flicker Induced by Large WT’, IEEE Transac-
tions on Energy Conversion 14(3) 743–748.
[16] Sørensen, P., Tande, J. O., Søndergaard, L. M., Kledal, J. D. (1996) ‘Flicker Emission Levels from WT’,
Wind Engineering 20(1) 39–45.
[17] Walker, M. K. (1989) ‘Electric Utility Flicker Limitation’, in Proceedings of the Power Quality Conference,
Long Beach, California, October, pp. 602–613.
364 Practical Experience with Power Quality
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 18_CHA17 .3D – 365 – [365–382/18]
17.12.2004 10:39PM
17
Wind Power Forecast for the
German and Danish Networks
Bernhard Ernst
17.1 Introduction
Wind-generated power now constitutes a noticeable percentage of the total electrical
power consumed and in some utility areas it even exceeds the base load on the
network. This indicates that wind is becoming a major factor in electricity supply
and in balancing consumer demand with power production. A major barrier to the
integration of wind power into the grid is its variability. Because of its dependence on
the weather, the outp ut cannot be guaranteed at any particular time. This makes
planning the overall balance of the grid difficult and biases utilities against using wind
power. Accurate forecasting of power inputs from wind farms into the grid could
improve the image of wind power by reducing network operation issues cau sed by
fluctuating wind power.
The variab ility of wind power can also affect the price that is paid for wind-
generated electricity. Some countries give a subsidised price, higher than the pool
price, to wind generators. The trend, however, is towards deregulated electricity
markets. Under these regimes, producers may have to contract to provide firm power
and might be penalised if they overproduce or underproduce. This reduces the value of
the energy they sell which in turn reduces the incentive to invest in wi nd energy. By
aggregating the power output of wind farms and obtaining accurate predictions of the
power that will be fed into the grid it may be possible to raise the achieved price, and
this would make wind power a more favourable investment. Higher prices would also
mean that more sites will become feasible, as those with lower average wind speeds
would become economic.
Wind Power in Power Systems Edited by T. Ackermann
Ó 2005 John Wiley & Sons, Ltd ISBN: 0-470-85508-8 (HB)
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 18_CHA17 .3D – 366 – [365–382/18]
17.12.2004 10:39PM
The requirements regarding wind power prediction systems vary considerably
depending on the spec ific market and the distribution of wind turbines. The French
utility Electricite
´
de France (EDF), for instance, want s a day-ahead as well as a week-
ahead forecast. The value to be forecasted is the mean wind power production over
thirty minutes, on a national scale (France and other European countries) and on a
regional scale (Dispower, 20 02). The British New Electricity Trading Arrangement
(NETA) includes completely different requirements. The spot market closes only one
hour before the time of delivery. However, generators find two-hour predictions most
useful because of the work that has to be done prior to the de livery. In Denmark and
Germany, predictions are used mainly for the day-ahead market, which closes at
noon in Denmark (i.e. up to 36 hours before time of delivery) and at 3 p.m.
in Germany (up to 33 hours before the time in question). Hence, the requirements for
wind power forecast systems are usually set by market operation constraints, not by
technical or physical constraints. The hydro-based system in Scandinavia, for instance,
technically does not require a 36-hour forecast as the hydro system can adjust
generation in a fraction of an hour. However, market operation requires a 36-hour
forecast.
In addition to the different timespans that forecasts cover, different spatial reso-
lutions are required. In Denmark and in Germany, for instance, wind turbines are
spread all over the country. Therefore, the wind power is forecasted mainly for large
areas (see also Chapters 2, 10 and 11). In the USA and in Spain, the installed wind
power is limited to a number of large wind farms. This requires a forecast for single
wind farms instead of wide areas. However, most of the prediction systems tend to
cover all possible needs.
17.2 Current Development and Use of Wind Power Prediction Tools
Meteorological and research and development (R&D) institutes initiated and devel-
oped the short-term prediction of wind power. The Risø National Laboratory and the
Technical University of Denmark were first in this area. They have used detailed,
area-specific, three-dimensional weather models and have worked with numerical
weather prediction (NWP) models, such as HIRLAM (High Resolution Limited Area
Model), the UK MESO (UK Meteorological Office Meso-scale model) or the LM
(Lokal-Modell of the German Weather Service). These systems work like a weather
forecast, predicting wind speeds and directions for all the wind farms in a given area.
On this basis, they calculate the output power using physical equations such as in
WAsP (Wind Atlas Analysis and Application Program, http://www.wasp.dk) or
similar programs. The models make it possible to produce predictions for up to 48
hours ahead. They have been designed primarily to provide information for network
companies to be able to operate their power systems. Similar projects have been
carried out
.
in Norway at the Norwegian Meteorological Institute;
.
in the USA at Wind Economics and Technology (WECTEC) and at TrueWind
Solutions;
366 The German and Danish Networks
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 18_CHA17 .3D – 367 – [365–382/18]
17.12.2004 10:39PM
.
in Germany at the Institute of Solar Energy Supply Systems (ISET), at Oldenburg
University and at Fachhochschule Magdeburg;
.
in Japan, at Mie University and the Toyohashi University of Technology;
.
in Ireland at the University College Cork.
Some of the systems use statistics, artificial neural networks (ANNs) or fuzzy logic
instead of physical equations. Their advantage over standard computing is that they are
able to ‘learn’ from experience. The disadvantage is that they need a large dataset to be
trained with be fore the system works properly.
In the following, several systems will be described in more detail. The main focus is
on two systems used by the transmission system operators (TSOs ) in Denmark and in
Germany. In Denmark, the Wind Power Prediction Tool (WPPT) was developed by
the University of Copenhagen and is used by the TSO in Western Denmark, Eltra,
whereas the TSO in Eastern Denmark Elkraft uses a similar model that was developed
in-house. In Germany, E.ON, RWE Net and VET use the Advanced Wind Power
Prediction Tool (AWPT), which was developed by ISET and predicts 95 % of German
wind power.
17.3 Current Wind Power Prediction Tools
17.3.1 Prediktor
The Prediktor system has been de veloped by Risø National Laboratory, Denmark, and
it uses, as far as possible, phy sical models. Large-scale flow is simulated by an NWP
model – HIRLAM in this case. The wind is transformed to the surface using the
geostrophical drag law and the logarithmic wind profile. When zooming in on the site,
more and more detail is required. Such de tails is provided by the Risø WAsP program.
WAsP takes into account local effects (lee from obstacles, the effect of roughness and
roughness changes, and speed-up or down of hills and valleys). The Risø PARK
program is used to include the shadowing effects of turbines in a wind farm. Finally,
two model output statistics (MOS) modules are used to take into account any effects not
modelled by the physical models an d general errors of the method.
The model runs twice a day generating 36-hour predictions for several wind farms in
the supply area. For each wind farm, power predictions can also be scaled up in order to
represent a region instead of only the wind farm. In order to evaluate the model, the
predictions are compared with those of the persistence model, which is a very simple
model stating that
Pðt þ lÞ¼PðtÞ; ð17:1Þ
where P(t) is the production at time t, and l is the ‘look-ahead time’. This model could
popularly be called the ‘what-you-see-is-what-you-get’ model. Despite its apparent
simplicity, this model describes the flow in the atmosphere rather well, because of the
characteristic time scale of weather systems; the weather in the afternoo n often is the
same as it was in the morning. For a short prediction horizon of only a few hours, this
model is hard to beat.
Wind Power in Power Systems 367
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 18_CHA17 .3D – 368 – [365–382/18]
17.12.2004 10:39PM
A comparison of the two models gives the following results (Landberg, 2000; Landberg
et al., 2000b):
.
The prediction model outperforms the persistence model after six hours.
.
The mean absolute error of the prediction model is around 15 % of the installed
capacity. A well-predicted wind farm has a scatter of as little as 10 %, and a not so
well predicted wind farm has a scatter of up to 20 %.
.
The performance of the prediction model deteriorates very gentl y – during 36 hours
the error increases by less than 10 %.
17.3.2 Wind Power Prediction Tool
WTTP implements another philosophy for predicting wind power. It was developed as a
cooperation between Eltra and Elsam and the Department of Informatics and Mathe-
matical Modelling (IMM) at the Technical University of Denmark (DTU). WPPT
applies statistical methods for predicting the wind power production in larger areas. It
uses online data that cover only a subset of the total population of wind turbines in the
area. The relevant area is divided into subareas, each covered by a reference wind farm.
Then local measurements of climatic variables as well as meteorological forecasts of
wind speed and direction are used to predict the wind power for periods of time ranging
from half an hour to 36 hours. The wind farm power prediction for each subarea is
subsequently upscaled to cover all wind turbines in the subarea; and then the predictions
for subareas are summarised to arrive at a prediction for the entire area.
Input data to WPPT are wind speed , wind direction, air temperature and the power
output at the reference wind farms. They are sampled as 5 minute mean values. In
addition to that, meteorological forecasts are used as an input to the models. The
forecasts are updated every 6 hours and cover a period of 48 hours ahead with a 1 hour
resolution. In general, WPPT uses statistical methods to determine the optimal weight
between online measurements and meteorological forecasted variables.
The first version of WPPT went into operation in October 1994. This version did not
use any NWP data. Predictions were based on the statistical analysis of measurements
from only seven wind farms. These predictions were not suitable for forecast horizons
that exceeded 8 hours:
.
because of the poor reliability of measurement equipment;
.
because too few wind farms were included;
.
because of a lack of meteorological forecasts.
A new project was started in 1996. Its aim was to obtain better predictions by
including more wind farms, by improving the reliability of the equipment and by getting
good meteorological forecasts. The ELTRA dispatch centres use this system with good
results. There, it provides forecasts for more than 2 GW installed wind power capacity.
Figure 17.1 shows the distribution of the error of 36 hour predictions for the year
2001. At that time, installed capacity was 1900 MW. Some 37 % of the errors are within
100 MW; large errors over 500 MW occur only 7 % of the time. Figure 17.2 shows the
368 The German and Danish Networks
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 18_CHA17 .3D – 369 – [365–382/18]
17.12.2004 10:39PM
40
35
30
25
20
15
10
5
0
Prediction error (MW)
Percentage of time
–1300
1300
–1100
1100
–900
900
–700
700
–500
500
–300
300
–100
100
Error, 36 h
Figure 17.1 Distribution of prediction errors for 1900 MW of installed wind power with a
prediction horizon of 36 hours
Source: Holttinen, Nielsen and Giebel, 2002. (Reproduced by permission of KTH, Sweden.)
Prediction horizon (hours)
Absolute error (%)
100
90
80
70
60
50
40
30
20
10
0
0 6 12 18 24 30 36
WPPT prediction
Persistence
Figure 17.2 The total absolute prediction error during one year for different prediction horizons,
as a percentage of the total production in 2001. Note: WPPT ¼ Wind Power Prediction Tool
Source: Holttinen, Nielsen and Giebel, 2002. (Reproduced by permission of KTH, Sweden.)
Wind Power in Power Systems 369
//INTEGRAS/KCG/P AGIN ATION/ WILEY /WPS /FINALS_1 4-12- 04/0470855088_ 18_CHA17 .3D – 370 – [365–382/18]
17.12.2004 10:39PM
total prediction error depending on the forecast horizon. For forecast horizons of more
than 3 hours, WPPT outperforms the persistence model. The graphs were produced with
the 1997 version of WPPT. In 2003, a new, improved, version was installed, (Holttinen,
Nielsen and Giebel, 2002; Nielsen, Madsen and Tofting, 2000).
17.3.3 Zephyr
In Denmark, a new program for short-term predictions of wind energy is being devel-
oped. Risø National Laboratory and IMM jointly develop the system, and all Danish
utilities will use the system. The new software is called Zephyr and, being based on
Java2, it offers a high flexibility. New prediction models have beco me possible because
new input data are available. Zephyr merges the two Danish models Prediktor and
WPPT. This combination will ensure that there are reliable forecasts available for all
prediction horizons, from short-range (0–9 hours) to long-range (36–48 hours) predic-
tions, as IMM uses online data and advanced statistical methods , which provide good
results for the short-range prediction horizon. In addition to that, the use of meteoro-
logical models such as the HIRLAM model of the Danish Meteorological Institute
benefits the long-term forecasts. An intelligent weighting of the results of both modelling
branches then gives the result for the total supply area of the customer (Landberg,
2000a).
17.3.4 Previento
Previento is the name of a forecast system, that has been developed recently at
Oldenburg University. It provides power predictions for a larger area (e.g. the whol e
of Germany) for the following two days. Its approach is similar to that of the Prediktor
system. The German version uses data of the German Weather Service instead of
HIRLAM data. Furthermore, it provides an estimation of the possible error depending
on the weather situation (Beyer et al., 1999; Focken, Lange and Waldl, 2001).
17.3.5 eWind
The eWind
TM
system was developed by True Wind Solutions, USA, and runs now at
Southern California Edison (SCE) Company, with about 1000 MW of installed wind
power. The system is composed of four basic components:
.
a set of high-resolution three-dimensional physics-based atmospheric numerical models;
.
adaptive statistical models;
.
plant output models;
.
a forecast delivery system.
The physics-based atmospheric models are a set of mathematical equations that repre-
sent the basic physical principles of conservation of mass, momentum and energy and the
equation of state for moist air. These models are similar to those used by operational
weather forecast centres throughout the world. However, the eWind
TM
models are run
370 The German and Danish Networks