W I ND E N G I N E E R I N G V O L U M E 26, N O .

4 , 20 0 2 PP 19 1–20 8 19 1

Dynamic modelling of wind farm grid interaction

Anca D. Hansen1, Poul Sørensen1, Frede Blaabjerg2 and John Becho3
1 Risø National Laboratory, Wind Energy Department, DK-40 0 0 Roskilde, Denmark;
emails <anca.daniela.hansen@risoe.dk>, <poul.soerensen@risoe.dk>
2 Aalborg University, Pontoppidanstræde 10 1, P.O. Box 49, 9220 Aalborg East, Denmark, DK;
email < fbl@iet.auc.dk>
3 NEG-Micon Control Systems, Italiensvej 1-5, DK-8450 Hammel; email < jbe@dancontrol-eng.dk>

ABSTRACT
This paper describes a dy namic model of a wind farm and its nearest utility grid. It is
intended to use this model in studies addressing the dy namic interaction betw een a wind
farm and a power sy stem , both during norm al operation of the wind farm and during
transient g rid fault events. T he m odel comprises the substation w here the wind farm is
connected, the internal power collection sy stem of the wind farm , the electrical, mechanical
and aerody namic models for the wind turbines, and a wind m odel. T he integ rated m odel is
built to enable the assessm ent of power quality and control stra tegies. It is implem ented in
the comm ercial dedicated power sy stem sim ulation tool D IgSILE NT.

1 INTRODUCTION
The motivation for this investig ation is the ever-increasing w ind energy penetration into
power network s. In recent years the trend has been moved from installations w ith a few w ind
turbines to the planning of larg e w ind farms w ith more than hundreds of MW of capacity. This
increased and concentrated penetration mak es the power network m ore dependent on, and
v ulnerable to, the w ind energ y production. This situation means that future wind farm s must
be a ble to replace conventional power stations, and thus be active controllable elements in the
power supply network . In other w ords, w ind farm s must develop power plant characteristics
(Sørensen P. et al., 2000). The tw o utilities responsible in D enmark for power supply netw ork s,
Eltra and Elk raft Sy stem, have issued requirements (Eltra, 2000) that focus on the influence of
wind farms on grid stability and power quality, a nd on the control capabilities of wind farms.
A nother consequence of the increased future size of wind farm s is that the large wind
farm s will be connected directly to the high voltage tra nsmission grid. Until now, w ind
turbines and w ind farm s have been connected to the distribution sy stem , w hich ty pically has
either 10/ 20 k V or 50/ 60 k V grids. Therefore, the m ain focus has been on the influence of the
wind farm s on the power quality of the distribution sy stem . F or exa mple in D enmark , this has
been reg ulated by the D anish Utilities R esearch Institute (D EF U ) requirements for g rid
connection of wind turbines to the distribution sy stem (D EF U K R111, 1998). H owever, the
transmission sy stem operators in D enmark now issue m ore strict connection requirements for
larg e wind farms if they are connected directly to the transmission sy stem . Moreover, national
standards for power quality of wind turbines have recently been supplem ented by a new
standard for measurement and assessm ent of power quality of grid connected w ind turbines,
namely (IEC 61400-21, 2001).
Having in mind all these aspects, it is important to appreciate the information provided by
models that simulate the dy namic intera ction between a wind farm a nd a power sy stem . Such
192 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

developed models enable both the w ind farm investors and the utility grid technical staf f, to
perform the necessa ry preliminary studies before connecting w ind farm s to the g rid.
Simulation of the wind farm interaction w ith the g rid may thus provide valuable information
and m ay even lower the overall g rid connection costs. The present w ork describes the first
steps to provide such m odels. The described w ind farm sim ulation m odel is the first of its kind
and represents a complex tool in the prediction of the influence of w ind farm s on the power
quality. The model can be used and even further ex tended in order to study different aspects
related to the connection of the w ind farm s to the g rid, i.e. control strategies, connection to
weak g rids, H V D C connections a nd storage sy stems.
This paper presents the m ain result of a national D anish resea rch project, nam ely a w ind
farm m odel capable of simulating the intera ction between a wind farm and its power network ,
both during norm al operation and during transient grid fault events (Sørensen P. et al., 2001a).
The model is developed in the power sy stem simulation tool D IgSILENT, w hich provides both
an ex tensive library for g rid components a nd a dy namic simulation languag e (D SL) for the
modelling of each w ind turbine component. The case study for this research is the 12 MW w ind
farm installed in Hag esholm, D enmark , w hich consists of six NM-2000 constant speed w ind
turbines from NEG -Micon, each of 2 MW rated power with active stall control. H agesholm
wind farm is connected to a strong grid compared to the w ind farm power capacity. This small-
scale case study establishes the first steps toward the larg e-scale case study, i.e. a 150 MW w ind
farm connected to the transmission network .
The w hole w ind farm m odel comprises a g rid model, six almost identical wind turbine
models and a wind speed model, as follow s:
l the grid m odel includes the electric components in the w ind farm power collection
sy stem a nd the substation, and uses equivalents for the remaining g rid.
l the individual wind turbine m odel includes electrical, m echanical a nd aerody namic
submodels:
– the electrical subm odel consists of an induction g enerator, a soft-starter, a
capacitor bank for reactive power compensation and a step-up transformer.
– the m echanical subm odel describes the dy namics of the drive train.
– the aerody namic subm odel is a standard C p based m odel, ex tended with dy namic
stall effects.
l the wind speed m odel outputs the wind speeds for each w ind turbine in the w ind
farm , taking into account the coherence and the effects of w ind va riations in the
rotor plane (e.g. 3p fluctuations due to rotationally sa mpled turbulence and tower
shadow of the 3-bladed turbine).
The simulation model is validated in the first place, for fault-free operation, based on power
quality m easurem ent data on a wind turbine from Hag esholm wind farm , and the results are
promising.

2 MODELLING TOOL
Computer models of power sy stems are w idely used by power sy stem utilities to study load
flow, steady state voltage stability and dy namic behaviour of power sy stem s. The models
presented in this paper are implem ented in the power sy stem sim ulation tool D IgSILENT,
because it provides the ability to sim ulate load flow, R MS fluctuations and transient events in
the sam e software environment.
D IgSILENT provides a comprehensive library of models for electrical components in the
W I ND E N G I N E E R I N G V O L U M E 26, N O . 4 , 20 0 2 19 3

power sy stem e.g. generators, motors, power plant controllers, dy namic loads and various
passive netw ork elem ents, such as lines, transformers, static loads and shunts. In the present
w ork , the g rid m odel and the electrical components of the w ind turbine a re built w ith
standard component m odels from the D IgSILENT library. The m odels of the w ind speed, of the
mechanical, of the aerodynamic a nd control parts of the w ind turbines, are im plemented in
the dy namic simulation languag e D SL of D IgSILENT. D SL allow s the user to implem ent
specific m odels that are not standard in D IgSILENT library, and thus to create ow n developed
blocks, either as modifications of existing m odels or a s completely new models. These
developed m odels a re gathered in a personal library, w hich can be used easily for further
modelling of other w ind farm s with different w ind turbines.
D IgSILENT provides models w ith different detailing levels in a very w ell structured w ay. It
combines the traditional transient EMT (electromag netic transient) simulation tool for power
sy stem s with R MS sim ulations of longer-term dy namics. D epending on the g oal of the
analy sis, it is possible to select the appropriate detailed models, by choosing the ty pe of
simulation m ethod. The R MS simulations are based on electromechanical transient models,
w hich are more simplified models than the electromag netic transient models used in EMT
simulations. R MS simulations are m ore appropriate for long simulation periods w ithout
transients, a s in m ost studies of pow er quality and control issues. On the other hand,
simulations of instantaneous EMT values w ith detailed m odels are required for reliable
simulations of the behaviour during grid faults.

3 GRID MODELLING
Hag esholm w ind farm consists of six NM 2000/ 72 turbines from NEG -Micon, each of 2 MW
rated power w ith a ctive stall control. The wind farm is connected via two feeders to the 10 k V
busbar in 50/ 10k V substation in G revinge, w hich is located 5 k m aw ay from the w ind farm .
The netw ork layout of Ha gesholm wind farm is sk etched in Figure 1. It illustrates how these
six wind turbines are distributed in two g roups in their connection to the substation. Each of
the 6 wind turbines W T1, W T2,…, W T6 are connected to its ow n 10k V terminals called
“W T1_10k V ”,…, “W T 6 _10k V ”, respectively. E a ch terminal represents the w ind turbine
terminal to the g rid. A back up line, normally disconnected, is also installed between the ends
of the two lines. Tw o identical 16MVA 50/ 10k V tra nsformers are installed in G reving e
substation, of w hich only one (TR1-50/ 10k V in Figure 1) is connected in normal operation
conditions. The other transform er (disconnected) serves as a backup.
Both the connection of the wind farm to the substation a nd the substation itself are
modelled by the actual physical components (transformers, line, load, busbar), w hile a
simplified equivalent Thevenin is used to model the remaining power sy stem (50 k V g rid).
This is a fair approximation for power quality studies, a s the g rid is very strong compared to
the power capacity of the w hole w ind farm .
The load feeders, w hich are also connected to the G revinge substation, a re modelled in the
first place as a single, g eneral load connected to the substation directly to the 10k V busbar of
the substation – see Figure 1.

4 WIND TURBINE MODELLING

The w ind turbine model includes electrical, mechanical and aerodynamic m odels (Sørensen
P. et al., 2001a ). Each of the six w ind turbines of H agesholm wind farm has its ow n m odel, w hich
enables simulation of the interaction between the w ind turbines in the wind farm . The model
also mak es it possible to simulate with different genera tor ty pes on the individual w ind
194 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

turbines, as it is the case in H ag esholm. Besides of different g enerator parameters, the w ind
turbine m odels are identical.
Figure 2 illustrates the overall structure of the w ind turbine model. It consists of an
aerodynamic, a mechanical and an electrical m odel, with an overall control sy stem model.
The w ind m odel provides an equivalent w ind speed n eq
as input to the aerodynamic m odel.
The equivalent w ind speed n eq (see Section 5) tak es into account the rotational turbulence, the
tower shadow and the va riations in the w hole wind speed field over the rotor disk , by using the
information from the mechanical sy stem about the turbine rotor position u wtr .
Besides the
equivalent w ind speed veq, the pitch angle u pitch from the control sy stem and the w ind turbine
rotor speed v wtr from the mechanical model constitute the inputs of the aerody namic model.
The output of the aerody namic m odel is the aerody namic torque T ae , w hich is input for the
mechanical m odel together w ith the g enerator speed v gen . The m echanical model delivers the
mechanical turbine power P t to the generatow r model. The electrical m odel interfaces with
the power sy stem by the voltag es U wtt and currents Iw tt on the (10 k V ) wind turbine terminals.
The control sy stem uses as inputs the following signals: 1) the hub w ind speed n hu b ( fixed-point
wind speed), 2) the g enerator speed v gen a nd, 3) the active power P m s and reactive power Q m s
at the main switch, representing the m easured voltag es a nd currents of the control sy stem.
The control sy stem has as outputs: 1) the pitch angle u pitch to the aerody namic model, 2) the
soft-starter angle a ss and, 3) the capacitor bank sw itch signals SC to the electrical m odel.

Figure 1 Single line diagram of the Hagesholm wind farm grid in DIgSILENT
W I ND E N G I N E E R I N G V O L U M E 26, N O . 4 , 20 0 2 195

4.1 Aerodynamic model

Standard aerody namic program s ty pically use blade elem ent methods and therefore they
require considera ble computation time. This is not a dequate to sim ulate a wind farm with
sev eral w ind turbines. Therefore the present a erody namic m odel is based on the
aerodynamic efficiency C p( l ,u pitch ), w hich is calculated by a standard aerodynamic progra m.
This sim plification is a cceptable a s long as only the ef fect on the pow er, nam ely the
aerodynamic torque T ae, is tak en into account.
F or a given rotor, C p( l ,u pitch ) depends on the pitch angle u pitch and on the tip speed ratio l =
Rv w tr / n eq . The aerody namic efficiency, usually tabled as a m atrix in the a erody namic
program , is used to determine the quasi static aerody namic power P ae developed on the main
shaft of a w ind turbine w ith rotor radius R a t a w ind speed n eq a nd air density r :

2r p R n
Pae = – eq Cp(u pitch,l
1 2 3 ) (1)

Once the a erody na mic power is determined, the a erody na mic torque T ae can be
calculated directly a ccording to:
Pae p
Tae = v = r R3n 2C
p(u pitch,l ) (2)
wtr 2l eq

The simplification of using a steady state aerody namic efficiency, instead of a dy namic
aerodynamic efficiency, in Equation (1) corresponds to the steady state a ero loads. This
simplification underestimates the actual power fluctuations in the stall region. F or the sak e of
simplicity, from now on, the steady state a erodynamic efficiency is denoted by C pstatic , w hile
the dy namic aerody namic efficiency is referred by C pdy namic . Figure 3 illustrates the power
curve of a w ind turbine and how larg e the power fluctuations are, as caused by wind speed
fluctuations for two situations:low wind speed region (u < 8 m/ s – case 1) and high w ind speed
region (u > 8 m/ s – case 2). F or low wind speeds (case 1) there is no significant difference
between the static power P static and the dy namic power P dynamic , because the a mplification
factors of the fluctuations for both steady state a nd dy namic state are similar. This is not the
case for the high w ind speeds ( case 2) in the stall region, w here, due to the dy namic stall
effects, fluctuations in wind speed, produce larg er power fluctuations.

Figure 2 General structure of the wind turbine model

196 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

The D IgSILENT aerodynamic m odel is improved w ith a model for the dy namic stall effects.
The applied m odel for the dy namic stall is an enhancement of the method described in (Øye
& S., 1991). This former method sim ulates the dy namic stall as time lag of separation and it
im plements the time lag directly on the lift coefficients. The present aerodynamic m odel,
im plemented in D IgSILENT, is based on the aerodynamic efficiency coefficient C p, instead of
the lift coefficients C L . The dy namic aerody namic efficiency, C pdynamic , is determined as based
on the static aerodynamic efficiency, C pstatic , and two a dditional aerody namic efficiencies
concerning the w ind profile around the w ing, i.e. one for attached flow and the other for
separated flow. The attached flow corresponds to the steady state flow at low angles of attack,
w hile the separated flow corresponds to the steady state flow a t high angles of attack (Øye &
S., 1991). Thus, for a certain pitch angle, u pitch ,
a nd tip speed ratio, l , the three aerody namic
coefficient tables provide respectively the values for C pattached , C pstatic and C psepar ated . The static
value of the separation ratio f st is computed as illustrated in Fig ure 4. Then, the dy namic
separation ra tio f is obtained by using a low pass filter, w ith the tim e constant (la g )
approximated by t = 4/ u 0 , w here u 0 is the m ean w ind speed.

Figure 3 Power fluctuations in the stall region with and without dynamic stall

Figure 4 DIgSILENT block diagram for dy namic stall

W I ND E N G I N E E R I N G V O L U M E 26, N O . 4 , 20 0 2 197

Finally, the dy namic value of the aerody namic coefficient is determined through the
interpolation:

Cpdynamic = f Cpattached + (1–f) Cpseparated (3)

This value is then used in the a erodynamic power ex pression (1) to estimate the power
fluctuations in the stall region.

4.2 Mechanical model

In the mechanical m odel, Figure 5, emphasis is put only on those parts of the dy namic
structure of the w ind turbine that contribute to the interaction with the grid. Therefore only
the drive train is considered in the first place, because this part of the wind turbine has the
most significant influence on the power fluctuations. The other parts of the w ind turbine
structure, e.g. tower and the flap bending modes, a re thus neglected.
The drive train converts the aerody namic torque T ae on the rotor into the torque on the
low speed shaft T l ss, w hich is scaled dow n through the gearbox to the torque on the high-speed
shaft T hss. The drive train model is a ty pical 2-m ass m odel, connected by a flexible low speed
shaft, w hich is m odelled by a stiffness k s and dam ping coefficient c s. The m asses used in the
model correspond to large turbine rotor inertia I wtr , representing the blades and the hub, and
a small inertia Igen representing the induction genera tor. Moreover, an ideal g ear w ith the
exchange ratio (1:h ) is included in the m odel. The high-speed shaft is a ssum ed to be stiff: T hss =
T l ss / h .
The g enera tor’s inertia Igen is implem ented in D IgSILENT as a part of the g enerator model,
w hile the remaining part of the mechanical model is described in a state space form in the
– –
dy namic simulation languag e D SL of D IgSILENT. The input va riables u , the output variables y
–
and the state variables x , respectively, are defined as follow s:

1 2 1 2
Thss u
u– = v1 2
k
gen , y– = u wtr , x– = u wtr
(4)
Tae v wtr v wt

w here u k is the angular difference between the two ends of the flexible shaft, i.e.

u
u k =u wtr –u lss =u wtr – h gen (5)

Figure 5 Drive train model

198 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

The state spa ce equations of the m echanical m odel are then:

1 2 1 2
0 0 1 – h 1– 0
–·
x= 0 0 1 ·x– + 0 0 ·u– (6)
– ks 0 – cs cs 1
Iwtr Iwtr h ·Iwtr Iwtr

1 2 1 2
ks cs c
s
—
h
0 —h
–—
h 2
0
– (7)
y= 0 1 0 ·x– + 0 0 ·u–

0 0 1 0 0

The dam ping coefficient c s is given by :

––––
cs = 2 j Ö ks Iwtr (8)

w here j is the damping ratio and can be ex pressed using the logarithmic decrem ent d s :

d s
j = (9)
Ö d 2 + 4p 2
s

The logarithmic decrem ent is the loga rithm of the ratio between the am plitude at the
beginning of the period and the am plitude at the end of the nex t period of the oscillation:

1 a(t+tp) 2
a(t)
d s = 1n (10)

w here a denotes the am plitude of the signal.

4.3 Electrical model

The electric design of each w ind turbine in H ag esholm is based on the ty pical “D anish”
concept, with tw o induction generators, a soft-starter, capacitor bank s for power factor
compensation and a step-up transform er – see Figure 6.

Figure 6 Simplified electrical scheme a wind turbine

W I ND E N G I N E E R I N G V O L U M E 26, N O . 4 , 20 0 2 199

Only one g enerator is connected to the 960V busbar, a t a time. This corresponds to the tw o
sets of w indings in the real generator that m ay operate w ith 4 or 6 poles.
D IgSILENT provides models as m odules for the electrical components of the w ind turbine,
nam ely g enera tor, soft-sta rter, capacitor bank s for rea ctive pow er compensa tion,
transform er, tower cable and busbar. The control sy stem is implem ented in D IgSILENT
dy namic simulation language D SL.
The electrical part of the w ind turbine W T1, im plemented in D IgSILENT, is illustrated in
Figure 7. It contains all the electrical components of the w ind turbine from the tower bottom
to the nacelle in the tower top. The 10k V terminal of the first wind turbine “W T1_10kV ” is
identical to the terminal with the sam e name in Figure 1. The scheme presented in Figure 7 is
“an ex tension” of the single line diagram , show n in Figure 1. The scheme focuses on the
electrical m odel of the w ind turbine. Terminals “W T1_10k V ” a nd “W T1_Top _10k V ”
correspond physically to the bottom a nd to the top of the w ind turbine, respectively. The 960
V busbar, with main sw itch MSP, the capacitor bank (C1_t1,…C10 _t1) and the soft-starter are
placed in the nacelle. A s the LV / MV step-up transform er “2-W inding Transformer _t1” is also
placed in the nacelle, the “Tower Ca ble _t1” is a m edium voltag e MV cable. The g enerator in the
NM-2000 w ind turbine is a squirrel cag e induction ty pe. Besides the electro-m ag netic
description, the generator model in D IgSILENT also contains the mechanical inertia of the
genera tor rotor, as it wa s mentioned in the previous section.
The cut-in of the g enerator to the g rid is performed by a soft-starter using firing angle
control, w hen the genera tor speed reaches the sy nchronous speed. W hen the soft-starter is
by passed, and thus the g enerator is connected, the power factor compensation is perform ed
by the capacitor bank. The capacitor bank consists of 10 capacitors controlled independently.
Based on the reactive power Q m s, m easured on the low voltag e side of the step-up
transform er at the main sw itch point MSP, the control sy stem close or open the individual
capacitors C1_t1,…, C10 _t, Fig ure 7. Figure 8 illustrates the D IgSILENT block diag ra m for the
capacitor bank controller. The diagram is g eneral for the sim ulation of capacitor bank
sw itchings, and consequently it can also be used to simulate other wind turbines. Only the
number of steps and averaging time m ay differ from one wind turbine ty pe to another.
The switching of capacitors is done based on the a verage value of the m easured reactive
power during a certain period of time. This a verage value is used to avoid rapid switchings of

Figure 7 Single line diagram of the electrical model of the wind turbine, implemented in DIgSILENT
20 0 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

the capacitors, due to the short variations in the reactive power, and thus, to stabilise the
sw itching and to protect the contacts. First a moving averag e of the rea ctive power is a pplied,
and then a sam ple-hold function is used to represent block averages. The sample-hold
function stores its sam pled input signal at the output until the nex t rising edge of the clock
signal appears. Such output a verage includes the closed loop effect of the rea ctive power
supply from capacitors, w hich are already connected at a certain mom ent. It is then used to
determine how m any a dditional capacitors should be connected or disconnected. The total
(open loop) number of capacitors, w hich are connected at each step, is obtained by a digital
integrator, im plemented in the block “Sum slot”, Figure 8. A t each time step, this block has
information about the closed loop reactive power and the m emorised number of capacitors
used previously. T he output of the digital integrator is then sent to the control block
“Switching Slot”, w hich opens and closes the connections of the capacitors.

5 WIND MODELLING
5.1 Introduction
A wind farm connected to the electrical g rid a ffects the supplied power quality, due to the
fluctuating character of its output power. The fluctuations are a consequence of the varia tions
of the w ind a nd they influence the control characteristics of the w ind farm. Therefore,
modelling the wind is essential in order obtain realistic sim ulations of the power fluctuations
during continuous operation of the wind farm.
The wind m odel used in this article is described in detail in (Sørensen P. et al., 2001b). The
emphasis here is laid on the structure and im plementation issues of the model. The main
advantag es of this wind model are the fast computation, reduced m em ory requirements and
the ease of use, either for varia ble or constant speed m odels. This w ind speed m odel is very
suitable for simultaneous simulation of a larg e number of w ind turbines, m aking it possible to
estima te efficiently the im pact of a larg e w ind farm on the power quality.
The wind m odel of each w ind turbine combines tw o ty pes of effects, namely deterministic
effects and stochastic effects. The deterministic part of the w ind m odel contains the m ean
wind speed and tower shadow variations. The stochastic part of the wind model includes the
park scale coherence 1 between the wind turbines in a w ind farm as w ell a s the effects of the

Figure 8 DIgSILENT block diagram for capacitor bank controller

1 The wind farm pow er production can be classified in tw o scale effects. The first scale depends on each
wind turbine dy namics and the second scale depends mainly on the interactions between wind turbines
in the park as well as the coherence of the turbulence. The turbulence between two positions is
correlated. The deg ree of correlation depends on the distance between the tw o positions, the orientation
of the wind and the turbulence intensity.
W I ND E N G I N E E R I N G V O L U M E 26, N O . 4 , 20 0 2 20 1

rotational turbulence, namely that turbulence seen by the rotating w ind turbine blades. Only
the horizontal component of the w ind speed is considered in the m odel, as this component has
the m ost dominating influence on the aero loads on w ind turbine. The park scale coherence
ensures m ore accurate predictions of the m axim um output power from the wind farm,
especially in the case w hen the wind turbines are sited very close together or w hen the w ind
blow s in a direction close to the orientation of a line of w ind turbines. Each w ind turbine in the
wind farm ex periences its local version of the “global” wind turbine. The rotational turbulence
and the tower sha dow are included in the m odel because they cause fluctuations in the
individual turbine power. F or the 3-bladed turbines, this fluctuation is at three times the
rotational frequency (3p), and is the main cause of flick er in the turbine power during
continuous opera tion (Sørensen P., 1995). The tower shadow is modelled a s a 3p fluctuation
with constant a mplitude, w hereas the rotational sampled turbulence is m odelled a s a 3p
fluctuation with variable amplitude (Sørensen P. et al., 2000).
The structure of the wind model is illustrated in Figure 9. It is built a s a two-step model. The
first step is the park scale wind m odel, w hich sim ulates the fixed point (hub height) wind
speed n hu b,i a t each w ind turbine, taking into account the park scale coherence. The second
step is the rotor w ind m odel, w hich for each w ind turbine i, provides a n equivalent wind speed
n eq,i to the aerody namic model – see Figure 2. The park scale model provides w ind speed time
series at each w ind turbine hub n hu b,i for the rotor wind m odel. The rotor wind m odels add the
effect of the integration of the fixed-point fixed speed n hub,i
over the w hole rotor, the effect of
the rotational turbulence and the effect of the tower shadow. The wind shear is not included in
the rotor w ind model, a s it only has a sm all influence on the power fluctuations.
A s illustrated in Figure 9, the w ind model provides an equivalent w ind speed n eq,i
for each
wind turbine i, and not an equivalent wind speed for the w hole wind farm .

5.2 Park scale wind model

The park scale coherence model is a stochastic m odel, w hich sim ulates w ind speeds with the
sa m e stochastic chara cteristics a s m ea sured w ind speeds a t fix ed points. These
characteristics are power spectral densities (PSD s), w hich describe the frequency content of

Figure 9 Simplified scheme of the wind models for a wind farm

20 2 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

the fluctuations, and the coherence functions, w hich describe the coherence between w ind
speeds in different points (i.e. a t the w ind turbines) in the frequency domain.
The park scale w ind model is im plemented in an ex ternal program PA R K W IND, developed
at Risø National Laboratory. It g enerates a file with w ind speed time series n hub,1 , …n hub,n , w hich
are then read by D IgSILENT in the rotor w ind m odel of each wind turbine, see Figure 9. This
ex ternal calculation of the hub wind speed in PA R K W IND is permitted because it is assum ed
that the hub wind speed is independent of the w ind farm opera tion. The hub w ind speed is
considered to be the w ind speed at the hub height of the wind turbine, i.e. as if the wind turbine
had not been erected. The present version of PA R K W IND does not yet include the effects of
’w a k es’ in the w ind farm, but m odifications on the m ean wind speed a nd turbulence intensity
according to (Jensen N.O., 1983), (F randsen S. et al., 2001) w ill be directly included in the nex t
version of PA RK W IND code. The PA R K W IND m ethod is capable of sim ulating w ind speed
with power spectra of either the K aimal (K aimal J.C. et al., 1972) or Højstrup ty pe (H øjstrup J.
et al., 1990). A ny coherence function is sim ple to use inside the PA R K W IND program
(Sørensen P. et al., 2001a). D avenport ty pe coherence (D avenport A .G ., 2002) and the decay
factors recommended by ( Schlez W. & Infield D ., 1998) are used as defaults in the program.

5.3 Rotor wind model

A s m entioned before, the conventional simulation program used for calculation of structural
loads on wind turbines, simulates the w ind speed at a large number of points in the rotor plane.
Then it interpola tes the w ind speed betw een these points to provide the w ind speed
distribution along the blades. Finally, blade element iteration is applied to a number of sections
along each blade. This is a relatively time-consuming m ethod, and thus difficult to use for a
simulation of hundreds of wind turbines.

Figure 10 Rotor wind model in DIgSILENT

W I ND E N G I N E E R I N G V O L U M E 26, N O . 4 , 20 0 2 20 3

The idea of the rotor wind m odel is therefore to produce a single equivalent w ind speed n eq
for each w ind turbine, w hich is used as input to the aerody namic m odel. Figure 10 show s the
D IgSILENT im plementation of the rotor wind m odel for each wind turbine (Sørensen P. et al.,
2001a). The presence of 1p and/ or 2p is an indication of some asy m metry in the 3-bladed rotor.
W ithout loss of g enerality, it can be assum ed that the wind turbine rotor is sy m metrical and
therefore the harm onics, not multiples of 3, cancel out. Therefore they are not present in the
equivalent w ind speed model. F urthermore, in the present im plementation in D IgSILENT, it is
assum ed that the w ind turbine structure a cts lik e a low pass filter. T his reduces the
significance of the harmonics higher than 3p. Thus, the rotor w ind m odel contains, in the first
place, only the effects of zero a nd third order harmonics.
In the rotor w ind m odel, the hub w ind speed n hub is a veraged over the w hole rotor
(through the second order filter for zero harm onic – “A dmittance Filter 0”). The variations due
to the rotational turbulence and tower shadow in the wind speed field over the rotor disk are
tak en into account through a K aimal filter and an admittance filter for the third harmonic –
“K aimal Filter” and “A dmittance Filter 3”. The K aimal and A dmittance filters fitted by
(Langreder, 1996) are used.
The azim uth position of the turbine rotor u wtr
, used as input in the rotor w ind m odel,
provides information about the rotational sampling, w hich causes fluctuations in power with
three times the rotational frequency. The tower shadow is modelled as cosines w ith constant
amplitude and phase.

6 VALIDATION AND VERIFICATION OF THE MODEL

The validation of the developed model in D IgSILENT is perform ed for fault free operation and
it is based on power quality measurem ents for one of the wind turbines, assessed according to
(IEC 61400-21, 2001). In this paper, the model verification focuses on the power fluctuations and
the mechanical loads. Som e other verification results, dem onstrating the ability of the model
to predict the power quality of the wind turbines, are illustrated in (H ansen A .D . et al., 2001)
and in (Sørensen P. et al., 2001a).
Figure 11 show s the simulation result from D IgSILENT, for one w ind turbine, a t an averag e
wind speed of 9 m/ s and a turbulence intensity of 12%. A s ex pected, the variations in the w ind
are transposed w ithin the electrical power. Both the simulated equivalent wind speed and the
electrical power fluctuate with three times the rotational frequency (3p).

Figure 11 Simulated equivalent wind speed and simulated electrical power

20 4 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

A s both m easurements and sim ulation results are stochastic processes, an appropriate
way to compare them is to perform a power spectral a naly sis. Figure 12 illustrates the power
spectral density (PSD ) of the m easured a nd sim ulated power during continuous operation for
the ty pical 9 m/ s run. Comparing the simulated PSD to the m easured one, it is observed that
the low frequencies (below around 0.2 Hz) and the area around 3p frequency (0.9 H z) are
predicted quite w ell by the sim ulations.
A s only the effects of zero and third harm onic for turbulence and the third harmonic from
tower shadow are considered in the wind model, the 1p frequency (0.3 H z) is not included in the
model. It is only observed in the measurements and relates to some asy m metry in the rotor.
Figure 13 show s the PSD for the sim ulated and m easured torques for the sam e run as in
Figure 12. Tw o sim ulated torques are show n, namely the aerodynamic torque T ae and the low
speed torque T l ss (see Figure 5), together w ith the measured low speed torque. It is seen that
the torsional vibration m ode (flexibility) of the main shaft, w ith the eigenfrequency 0.7 H z,
provides an additional energy of T lss relative to T ae ; i.e. the PSD of T l ss comes closer to the PSD
of the measured torque than the PSD of T ae . This indicates that the mechanical m odel predicts
the shaft flexibility quite w ell. A significant 1p frequency is also present in the PSD of the
measured torque. Besides the asy m metry reason of the rotor, m entioned before, in the case of
the m easured torque, a further reason could be that the torque measurements from strain
gauges are very sensitive to ‘cross talk ’ from sha ft bending.
There are some comm on rem ark s reg arding both Figure 12 and Figure 13:
l There is slightly more 3p effect in the simulated power/ torque than in the measured
power/ torque, respectively. The spik e at the 3p frequency in the sim ulated power
and torque could indicate that the m odelled tower effect is overestimated. A nother
factor is that the result is very sensitive to the value of the torsional eigenfrequency
in the sim ulation, w hich is fairly close to the 3p frequency.
l The harm onics of higher orders than 3p, i.e. 6p (1.8 H z), 9p (2.7 H z) and 12p (3.6 H z),
visible in measurem ents, are not included in the model in the first place, since they

Figure 12 Power spectral density (PSD) of measured and simulated power fluctuations of wind turbine no. 1 at wind
speed 9 m/ s
W I ND E N G I N E E R I N G V O L U M E 26, N O . 4 , 20 0 2 20 5

are considered to have less influence on the power and since they increase the
simulation time. H owever, they can be easily included in the m odel, if this is desired.
l There exists a significant energy around 5 H z in the m easurements, but not in the
simulation. A study of measured m echanical load indicates that this frequency
could be due to som e kind of mechanical vibration m ode of the rotor together w ith
the tower (Sørensen P. et al., 2001a).
l The torsional vibration mode of the m ain shaft, w ith an eigenfrequency around 0.7
Hz, can be observed both in m easurements and sim ulations.
In order to illustrate the influence of the dy namic stall effect over the w hole wind speed
range, a sim ulation of a single w ind turbine is perform ed. In this ex ample, the equivalent w ind
speed is designed (not modelled by the rotor wind m odel) to have a sinusoidal varia tion with
a 3p frequency, amplitude of 2m/ s and a linear increasing m ean value, a s it is illustrated in
Figure 14. The simulation is performed w ith a constant pitch angle, w hich corresponds to 2
MW power at high w ind speeds.

Figure 13 Power spectral density (PSD) of measured and simulated torque fluctuations of wind turbine no.1 at mean
wind speed 9 m/ s

Figure 14 Designed equivalent wind speed – input for wind turbine simulation
20 6 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

Figure 15 show s the profile of the dy namic power curve, m odelled w ith a nd without
dy namic stall. The cyclic behaviour of the sinusoidal w ind speed input signal is tra nsferred
through the second order dy namic m odel of the transmission sy stem to other signals, such as
the aerody namic power, the a erodynamic torque a nd the rotor speed v w ith a corresponding
phase shift. The significantly increased am plitude of the dy namic power fluctuations at high
wind speeds illustrates that the dy namic stall effect is im portant at large w ind speeds, w hile at
sm all w ind speeds it is not relevant. The improved aerody namic m odel with the dy namic
effect is thus able to simulate the larger fluctuations in the power in the stall region. It is also
observed that even the power curve w ithout dy namic stall has some cyclic behaviour. This is
due to the sinusoidal wind speed input signal, w hich is transferred to the other state variables
through the model dy namics. Beside the sinusoidal w ind speed, the aerody namic efficiency
without dy namic stall, ex pressed a s a function of w ind speed, C p ( n ,u ), also influences the
calculation of the sm all cycles in the power curve w ithout dy namic stall.
Figure 16 illustrates the aerody namic efficiency a s a function of tip speed ratio, l = v R/n , for
the steady state, a ttached, separa ted and dy namic flow, in the conditions of the previous
simulation. It is observed that the steady state aerody namic efficiency, C pstatic , illustrated this
time as function of l , does not contain any cycles. This is because C pstatic is an algebraic function
of rotor speed v , wind speed n and pitch angle u . It is also noted that the steady state C pstatic
converg es to the attached C pattached aerodynamic efficiency at high l values (low w ind
speeds), w hile at sm all l values (larg e wind speeds), the steady state C pstatic converg es to the
separated C psepar ated . Moreover, at sm all w ind speeds, the steady state C pstatic aerody namic
efficiency a nd the dy namic aerody namic efficiency C pdy namic do not differ sig nificantly,
because at sm all w ind speeds the dy namic stall effect is not relevant. A t larg e w ind speeds
(sm all l values) the values of the dy namic a erodynamic efficiency C pdynamic are retarded with
the time constant t of the first order filter (see Figure 4), and then m ove toward the steady
state aerody namic efficiency C pstatic on an almost “parallel” curve with the a ttached and
separated efficiency curves. The calculated value of C pdy namic is interpolated between these
“parallel” curves, w hich correspond to different separation ratios f ( Sørensen P. et al., 2001a).
The dy namic value C pdynamic is the result of the first order filter, w hich physically corresponds
to the fact that, it remains for a time lag on a specific intermediary curve. Thus, at larg e w ind
speeds, the va riation of C pdynamic is different from that of C pstatic .

Figure 15 Power curve with and without dy namic stall effect

W I ND E N G I N E E R I N G V O L U M E 26, N O . 4 , 20 0 2 20 7

Figure 17 illustrates the PSD analy sis for both measured and sim ulated power fluctuations
at a large mean w ind speed of 14 m/ s. Tw o sim ulations a re performed, one with and one
without the dy namic stall model. It is observed that the dy namic stall model im proves the
simulations slig htly, even though the PSD of the sim ulation without the dy namic stall model
looks acceptable.

Figure 16 Steady state, attached, separated and dy namic aerodynamic efficiency as a function of l

Figure 17 PSD of measured and simulated power fluctuations of wind turbine no. 1 at wind speed 14 m/ s
20 8 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

However, a m ore detailed impression on the character of the fluctuations is achieved on

the time series, illustrated in Fig ure 18. Tw o sim ulations are done w ith identical wind speed
input. It should be noticed that the sim ulated and the measured time series do not have the
sam e wind speed time series, so they cannot be compared step by step. The idea of the
comparison with the m easured time series is just to show how larg e the power fluctuations are
in reality, as compared to those sim ulated. The character of the first 300 seconds of the
simulations is in good ag reement with the measurements (the sim ulations have a slow
va riation), w hile the character in the last 300 seconds of simulations is significantly different.
Here the simulations have very reduced slow fluctuation. The fast ( 3p) fluctuations in the
simulation w ithout dy namic stall are also strongly reduced due to the flat character of the
power curve in the stall region, as the w ind speed increases (see Figure 3). On the other hand,
the dy namic stall model improves the sim ulation of the fast fluctuations in the stall region.
However, it is noticed that the slow fluctuations are not sim ulated w ell enough. There are
tw o possible reasons for this. One reason can be that the wind model only includes the
horizontal component of the w ind speed. The transverse wind speed component is quite
im portant in the region of larg e w ind speeds as compared to the region of small w ind speeds,
due to: (1) the power fluctuations are m uch larg er at increased w ind speeds, (2) the efficiency
curve C p ( l , u pitch ) is very steep in that region. Studies of the slope of the efficiency curve in the
stall region have show n that a transverse component of the wind speed of 1 m/ s will increase
C p by 5%, corresponding to in this case to 100 k W (Sørensen P. et al., 2001a). A nother reason for
the absence of slow fluctuations in the sim ulations may be the uncertainty of the aerody namic
data.

Figure 18 Time series of measured and simulated power fluctuations of wind speed 14 m/ s
W I ND E N G I N E E R I N G V O L U M E 26, N O . 4 , 20 0 2 20 9

7 CONCLUSIONS
The present w ork provides a w ind farm dy namic model, as a first step towards the long-term
objective of developing tools for study and im provement of the dy namic interaction between
wind farms and power sy stems to w hich they are connected.
The w hole w ind farm dy namic model, w hich includes the m ain effects that contribute to
the fluctuation of the power from a w ind farm, has been developed and im plemented in the
power sy stem sim ulation tool D IgSILENT. It comprises the substation w here the wind farm is
connected, the internal power collection sy stem of the wind farm, the wind dy namics a nd the
wind turbines. T he individual wind turbine m odel contains the m echanical m odel, the
aerodynamic model (improved w ith a m odel for dy namic stall effects) and the electrical
model (w ith subm odels for induction g enerator, soft-starter, capacitor bank for reactive
power compensation and transform er).
The im plemented wind farm model can be easily ex tended to model other larg e w ind
farm s, w ith other wind turbines. It is built to enable a ssessm ent of power quality and control
strategies. The m odel has been validated for fault-free operation. The verification, based on
one w ind turbine, show s generally g ood agreement between simulations a nd measurements,
although the simulations a t larg er wind speeds seem to underestimate the power fluctuations.
A g reem ent can be improved by including the transverse component of the w ind speed in the
model. The im plemented model is able to simulate a w hole wind farm w ith respect to the
electrical power. The nex t steps are to enhance the model to predict the behaviour of the wind
farm in the event of g rid faults and to benchmark the model on data from a wind farm
connected to a transmission network .

8 ACKNOWLEDGEMENT
This w ork w as carried out by the W ind Energy D epartment a t Risø National Laboratory in co-
opera tion w ith A alborg U niv ersity a nd NE G -Micon C ontrol Sy stem s. T he a uthors
acknowledge the financial support to the project by the D anish Energy A g ency contract
#1363/ 00-0003. The co-operation w ith NegMicon Control Sy stem A / S has also been very
useful. Special thanks are given to the North-West Sealand Energ y Supply Company, NV E,
w hich provided g rid data and assisted in the power quality m easurements.

RE F E R E N C E S
D avenport A .G . (1961). T he spectru m of horizontal gu stiness near the grou nd in hig h winds.
Q.J .R.Meteorol.Soc., 87, 194 –211.
D EF U (1998). C onnection of wind turbines to low and m ediu m voltage netw or k s., R eport no.
K R111-E , Eltek nik k omiteen (1998-10-09).
Eltra (2000). Specifications for C onnecting W ind F arm s to the Transmissi on Netw or k. ELT
1999-411a, Eltra, http:/ / w w w.eltra .dk .
F randsen S., Thøgersen, & M.L. (2001). Integrated fatigue loading for wind turbines in wind farms
by combining ambient turbulence and wakes. W ind Engineering , vol. 23, pp. 327–340.
Hansen A .D ., Sørensen P., Janosi L., & Bech J. (2001). W ind farm m odelling for pow er quali ty .
Proc. IECON ‘ 01, held in D enver, USA ..
Højstrup J., Larsen S.E., & Madsen P.H . (1990). Pow er spectra of hori zontal wind com ponents in
the neu tral atm ospheric boundary layer. In Ninth Sy m posium of Turbulence and
Dif fusion, ed. N.O . Jensen, L. K ristiansen and S.E . La rsen.A m erican Meteorology
Society, pp. 305 –308.
210 D Y NA M I C M O D E L L I NG O F W I ND FA R M G R I D I N T E R A CT I O N

IEC 61400-21 (2001). W ind turbine generator sy stems – Part 21: Measu rement and assessm ent of
pow er quali ty char acteristics of g rid connected wind tu rbines. Final D raft Interna tional
Standard 88/ 144/ F D IS International Electrotechnical Commission, IEC 2001-07-01, Ed.1.
Jensen N.O. (1983). A note on wind generator i nteraction. Risø National Laboratory, Roskilde,
D enmark . Risø–M-2411.
K aimal J.C ., W y ngaa rd J.C ., Izumi Y., & Cote O.R. (1972). Spectral characteristi cs of surface
layer turbulence. Q.J.R .Meteorol.Soc., 98, pp 563 –598.
La ng reder,W . (19 96) . Model s for variabl e speed wind tu rbi n es. M.Sc. T hesis, C rest
Loughborough University a nd Risø National Laboratory.
Schlez W., & Infield D . (1998). Horizontal , tw o point coherence for separ ati ons g reater than the
m easurement height. Boundary -Layer Meteorology 87. K uvw er A cademic Publishers,
Netherlands, pp.459 –480.
Sørensen P. (1995). Methods for cal culation of the flicker contribu tion f rom wind turbines.
Risø National Laboratory, R oskilde, D enmark . Risø-I-939.
Sørensen P., Ha nsen A .D ., Ja nosi L., Bech J., & Bak -Jensen B. (2001a ). Sim ul ation of interaction
between wind farm and pow er sy stem . Risø National La boratory. Risø-R -1281.
Sørensen P., H ansen A .D ., & Rosas P.A .C. (2001b). Wind m odel s for prediction of pow er
flu ctu ations f rom wind far m s. Proc ‘The Fifth A sia -Pa cific Conference on W ind
Engineering (A PCW EV )’, K yoto, Japan. Journal of W ind Engineering no.89, pp. 9 –18.
Sørensen P., Bak -Jensen B., K ristiasen J., H ansen A .D ., Janosi L., & Bech J. (2000). Pow er pl ant
char acteristics of wind farm s. W ind Power for the 21st Century. Proceedings of the
International Conference held at K assel, G erm any 25–27 September.
Øye,& S. (1991). Dynamic stall – sim ulated as tim e lag of separation. Proceedings of the 4th
IEA Sy m posium on the A erody namics of W ind Turbines. McA nulty, K .F. (Ed.), R om e,
Italy.