Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 11, NO.

4, OCTOBER 2020 2889

On the Dynamic Modeling of Marine VSC-HVDC


Power Grids Including Offshore Wind Farms
Luis M. Castro and Enrique Acha

Abstract—This article presents a new and comprehensive model- premier onshore wind resource sites have already been used,
ing framework to carry out system-wide dynamic studies of DFIG- after more than 30 years of sustained wind farm construction and
based wind farms embedded in multi-terminal VSC-HVDC power commissioning, and the current push is for offshore wind farms.
grids. Contrary to existing, well-developed simulation frameworks
for similar studies, using electromagnetic transient (EMT) solu- This is particularly the case of countries with access to the North
tions, this is an RMS-type formulation which maintains a high- Sea, the Irish Sea and the Baltic Sea. The forecast is that the
degree of fidelity while enabling much faster steady-state and current estimated wind power production in the North Sea would
dynamic simulations than what it is possible to achieve with EMT increase between 7 to 15-fold by 2040 [2]. The UK alone has five
simulators. The new RMS modeling framework includes AC/DC wind farm developments in the North Sea, namely, Dogger Bank,
power grids of an arbitrary size, topology and number of offshore
VSC-connected wind farms. A simulation tool with such a high Hornsea, Norfolk, Firth of Forth and Moray Firth, with a total
degree of modeling versatility and numerical efficiency does not combined capacity of 18 GW. For instance, the Dogger Bank,
currently exist elsewhere. This has required the development, using the largest wind farm development, will have 1800 5-MW wind
first principles, of the RMS model of a DFIG with explicit repre- turbines [3]. The wind turbine technology selected for this site
sentation of all the dynamic effects relevant for dynamic problems is the so-called doubly fed induction generator (DFIG). These
of the electromechanical type as opposed to the study of very fast
EMT phenomena. All the control functions and parameters of the generators have continued to evolve rapidly, almost doubling the
rotor-side converter, the grid-side converter and the DC link, are 5 MW capacity; the largest commercial DFIG wind turbine has
accounted for in the new DFIG model. The prowess of the new reached the 9.5 MW capacity [4].
formulation is demonstrated using a six-terminal VSC-HVDC link Existing and planned wind farms in the North Sea have
with two VSC-connected 200-MW wind farms. The impact of the shown that for installations laying more than 70 km away
wind farms’ operation on both the DC grid and the AC grids is
assessed. The fidelity of the output results of the new simulation from the shore; their connection to the mainland grid is more
tool is compared against those of the EMT-type model implemented advantageous using VSC-based DC transmission links than
in Simscape Electrical of Simulink. The article shows that they AC transmission. As the number and ratings of wind farms
favorably compare with each other, with differences inferior to in the North Sea increases, it would make sense to develop
3%. The computational efficiency of the new dynamic modeling a fully-fledged multi-terminal VSC-HVDC grid. One that will
framework for HVDC-connected wind farms is unassailable.
serve not only the energy evacuation of the wind industry but
Index Terms—AC/DC networks, DFIG, power system dynamic also the supply requirements of the oil and gas rigs installations.
simulations, VSC-HVDC grids, offshore wind farms.
Indeed, the same arguments apply to other world regions with
I. INTRODUCTION similar conditions as the North Sea, such as the Gulf of Mexico
or the North Pacific Ocean. The development of this technology
IND power today represents a sound commercial activ-
W ity, showing a spiral growth. At the end of 2017, there
were 539.123 GW of installed wind power capacity worldwide
is progressing well; by 2020 there will be two multi-terminal
VSC-HVDC links in existence, both in China: the Nan’ao-3T
±160 kV and the Zhangbei-4T ±500 kV. Nonetheless, very
[1]. In this electrical power industry sector, China leads the much engineering and technology development is still required
way with 188.392 GW, followed by USA with 89.077 GW before the multi-terminal VSC-HVDC becomes a worldwide
and Germany with 56.132 GW. In Europe as a whole, the reality.
Today, the DFIG technology is the most popular in both
Manuscript received May 25, 2019; revised November 4, 2019 and February onshore and offshore wind farm installations. Over the past
8, 2020; accepted March 10, 2020. Date of publication March 16, 2020; date of
current version September 18, 2020. This work was supported by the National decade, various technical problems arising from their massive
Autonomous University of Mexico (UNAM) under program UNAM-DGAPA- integration into power grids have been identified and resolved.
PAPIIT-Project IA103919 (Programa de Apoyo a Proyectos de Investigación e However, there are still several outstanding issues that are yet
Innovación Tecnológica). Paper no. TSTE-00578-2019. (Corresponding author:
Luis M. Castro). to be dealt with satisfactorily, such as the reliable integration
Luis M. Castro is with the Department of Electrical Energy, National Au- of offshore wind farms that lie in deep waters, far away from
tonomous University of Mexico, Ciudad de México 06320, Mexico (e-mail: the shore [5]–[8]. Not surprisingly, great many research efforts
luismcastro@fi-b.unam.mx).
Enrique Acha is with the Department of Electrical Engineering, Tampere have been made toward developing advanced DFIG-based wind
University 33520, Tampere, Finland (e-mail: enrique.achadaza@tuni.fi). farm models to study their inherent electrical oscillations by
Color versions of one or more of the figures in this article are available online modal analysis [9], the mitigation of subsynchronous interac-
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSTE.2020.2980970 tions in DFIG-based wind farms [10], and the operation of

1949-3029 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
2890 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 11, NO. 4, OCTOBER 2020

DC-connected offshore wind farms to reduce output voltage


variations [11]. The same applies to the VSC-based HVDC
technology where effective formulations have been developed
to study transient and small signal stability issues [12]–[15].
However, when it comes to assessing practical power grids,
RMS-type models are predominantly used by transmission sys-
tems analysts to execute the copious amounts of dynamic simu-
lations that the power grid requires at the planning and operation
stages. In such application areas, the use of variable-speed wind Fig. 1. VSC unit with ancillary components.
generator models featured by very detailed switching-based
converter models, [9]–[11], [16], [17], are not popular since their
long execution times in utility-level power grids would tax un- equations representing the steady-state and dynamic behaviors
necessarily the analysts’ day-to-day job. It is in this tenor that this of the marine HVDC grids, including VSC-connected wind
paper aims to contribute, by introducing an efficient and versatile farms, in a unified frame-of-reference.
framework to carry out comprehensive assessments of future
power grids which will contain many both onshore and offshore
DFIG-based wind farms, with the latter requiring VSC-HVDC II. MODELING OF MULTI-TERMINAL VSC-HVDC LINKS FOR
transmission technology for power evacuation. The modeling DYNAMIC SIMULATIONS
framework put forward in this paper is a natural progression of System-wide dynamic analyses of AC networks may be an ar-
a recently published model of multi-terminal VSC-HVDC links duous task because repetitive simulations are needed to properly
for system-wide RMS-type dynamic simulations [18]. assess the impact of prospective contingencies on the dynamic
operation of the whole network. This is even more demanding for
VSC-based transmission systems due to the increased size of the
A. Scope and Contribution
resulting grid model which encompass the otherwise decoupled
This paper introduces several new modeling features and AC networks, the VSC stations and controls, and the DC power
control principles in the area of variable-speed wind turbines grid. Fig. 1 shows the main ancillary elements of a VSC unit
in connection with submarine wind farms. The new model- [18], [19].
ing framework enables efficient steady-state and dynamic as- Applying a nodal power injection formulation to the VSC
sessments of generic multi-terminal AC/DC grids with VSC- terminals, the nodal active and reactive power equations of
connected DFIG-based wind farms. Concerning the control the VSC and on-load tap changing (OLTC) transformer are
strategy of the VSC-HVDC grid, the formulation considers three computed as (1a)–(1b), respectively, where ma stands for the
different types of VSC models, namely, converters exerting DC RMS modulating variables, E is the DC bus voltage, Gsw
voltage regulation, converters fixing a power transfer and con- accounts for the converter switching losses and Beq enables
verters supplying power to passive grids. The three VSC models the converter reactive power injection, with Vac = ma · E · ejφ
seamlessly combine with those of the proposed VSC-connected and γ = θv − φ. The VSC injected powers at their AC and DC
DFIG-based wind farms. nodes, k and j, defining θkm = θk − θm and Vac = |Vac |, are
The derivation of the control features employed by VSC- as follows:
connected wind farms is emphasized in the paper, such as the 
frequency-balancing capabilities that ensures that the wind farm Pkcal = Vk2 Gkk + Vk Vm [Gkm cos θkm + Bkm sin θkm ],
frequency is always maintained close to its nominal value. This m∈k
is carried out by ensuring a continuous balance of the wind 
Pjcal = Ej2 Gjj + Ej En Gjn .
farm output power and the power injected into the DC grid. To
n∈j
conform to the actual operation of VSC, the developed dynamic
model contains two control degrees, a DC current controller and Pv = Vv2 Gph − Vac Vv [Gph cos γ + Bph sin γ]
a modulation ratio controller aiming at frequency regulation Qv = −Vv2 Bph − Vac Vv [Gph sin γ − Bph cos γ]
and AC voltage control, respectively. A key contribution of Pdcv = Vac
2
Gph − Vac Vv [Gph cos γ − Bph sin γ]
(1a)
this paper is the new RMS-type DFIG model whose stator + (Vac / m a )2 Gsw
and rotor circuits have been designed to seamlessly combine Qdcv = −Vac2
Bph − Vac Vv [−Gph sin γ − Bph cos γ]
with those of the back-to-back DC link, thus giving rise to a −Vac2
Beq
new and elegant model, following the modular principles of
a VSC-connected wind farm. The PWM-driven converters are Pvl = Vv2 Gltc − T Vv Vk [Gltc cos (θv − θk )
described in terms of basic electric circuit elements, such as + Bltc sin (θv − θk )]
an ideal tap-changing transformer, resistors, inductors and a Qvl = −Vv2 Bltc − T Vv Vk [Gltc sin (θv − θk )
shunt equivalent susceptance to account for the reactive power −Bltc cos (θv − θk )]
(1b)
behavior of the rotor-side converter and the grid-side converter. Pkl = Tv2 Vk2 Gltc − T Vk Vv [Gltc cos (θv − θk )
The new multi-offshore wind farm model is described in detail, +Bltc sin (θv − θk )]
emphasizing the numerical implementation that enables a reli- Qkl = −Tv2 Vk2 Bltc − T Vk Vv [Gltc sin (θv − θk )
able solution of the highly non-linear set of algebraic-differential −Bltc cos (θv − θk )]

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
CASTRO AND ACHA: ON THE DYNAMIC MODELING OF MARINE VSC-HVDC POWER GRIDS INCLUDING OFFSHORE WIND FARMS 2891

TABLE I grids, these are taken to represent two-level, three-phase con-


TYPES OF VSC AND THEIR INVOLVED VARIABLES
verters. However, the proposed modeling framework can be
reformulated for positive-sequence dynamic simulations with
MMC-HVDC systems. But it would be necessary to determine in
advance the new AC-to-DC voltage relationship of the selected
multilevel converter, Vac = k · m · Edc · ejφ , specifically the
product k·m that represents the modular multilevel arrangement
These equations are appropriately used to formulate the and modulation strategy of the VSC station. Similarly, the DC
VSC-based AC/DC grid model, one featured by three types of voltage dynamics of MMCs can be different from those of two-
converters and whose involved variables are shown in Table I: level converters [20], [21]. Therefore, the definition of suitable
(i) VSCSlack – converters controlling their DC bus voltage, differential equations related to those dynamics to be resolved
Ej = Enom ; (ii) VSCP sch – converters regulating their power within this dynamic modeling framework is key for obtaining
flow, Pdcv = Psch ; (iii) VSCP ass – power converters feeding realistic results [21], [22]. The reader may refer to other relevant
into passive grids with no generation of their own. works related to the modelling of MMCs focusing on the analysis
Following a generalized modeling approach, an arbitrary DC of their internal dynamics, such as [23]–[25]. Having defined the
network formed by a multi-terminal arrangement of m VSCSlack AC-to-DC voltage relationship and the differential equations as-
converters, n VSCP sch converters and r VSCP ass converters is sociated with the internal energy dynamics of MMC units, their
denoted by FHVDC = −JMT-HVDC ·ΔΦHVDC , Eq. (2) shown power flow equations can be derived, as in [18], [19]. In turn,
at the bottom of this page, where the implicit trapezoidal inte- these can be used for positive-sequence dynamic simulations
gration method and the Newton-Raphson algorithm are selected of MMC-HVDC power grids. This modeling flexibility is the
to simultaneously solve the discretized differential equations of main reason why this platform is selected to incorporate VSC-
the VSC controls and the algebraic mismatch equations, at each connected offshore DFIG-based wind farm dynamic models, as
iteration i and time t. Hence, matrices JSlackm contain the first- shown in subsequent sections.
order partial derivatives of the m VSCSlack mismatch equations,
FSlackm , with respect to variables ΔΦSlackm ; JPschn represent
the Jacobian matrices resulting from the n VSCP sch units whose
derivatives are with respect to variables ΔΦPschn ; the first- III. DFIG-BASED WIND TURBINE MODEL
order partial derivatives resulting from the r VSCP ass mismatch
equations, FPassr , are contained in matrices JPassr . Similarly, One of the big challenges in today’s dynamic power sys-
the DC grid nodal mismatch equations and their corresponding tem analyses is the very long simulation times accrued when
derivatives are accommodated in FDC and JDC , respectively, repetitive studies are required, in a range of power system
where vector ΔΦDC contains the DC nodal voltages. Finally, applications. In this sense, the use of a switching-based modeling
the cross terms JSdc , JPsdc , JPadc , encompass the mutual approach to represent each VSC of the back-to-back converter
derivatives between the VSC mismatch equations with respect of a DFIG, where the PWM pattern is fully emulated, would
DC voltages, and vice versa for terms JdcS , JdcPs , JdcPa . The hinder the possibility of carrying out dynamic simulations of
control loops of the converters are reported in the Appendix. real-scale systems incorporating a multi-terminal VSC-HVDC
grid and DFIG-type wind farms. Conversely, an RMS-based
modeling approach would offer a feasible solution to this prob-
A. A Brief Discussion on the Modeling of MMC-HVDC Links
lem where the computing times are considerably reduced, com-
The dynamic modeling framework, summarized by (1)–(2), pared to those obtained using an EMT-based solution approach
features the inclusion of several VSC-connected AC systems. (switching-based VSC models). With this goal in mind, this
As for the employed VSC models to form the arbitrary HVDC section focuses on deriving an RMS-type model of the B2B

⎛ ⎞(i,t) ⎛ ⎞(i,t) ⎛ ⎞(i,t)


FSlack1 JSlack1 · · · 0 ΔΦSlack1
⎜ ⎟ ⎜ .. .. .. [JSdc ] ⎟

..
⎟ ⎜ . 0 0 ⎟ ⎜ .. ⎟
⎜ . ⎟ ⎜ . . ⎟ ⎜ . ⎟
⎜ FSlackm ⎟ ⎜ 0 · · · J ⎟ ⎜ ⎟
⎜ ⎟ ⎜ Slackm ⎟ ⎜ ΔΦSlackm ⎟
⎜ FPsch1 ⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ J · · · 0 ⎟ ⎜ ΔΦPsch1 ⎟
⎟ ⎜ ⎟
Psch1
⎜ ⎟ ⎜ . . . ⎜ ⎟

..
⎟ ⎜ 0 .. .. .. 0 [JPsdc ] ⎟ ..
⎜ . ⎟ ⎜
= −⎜ ⎟ ⎜ . ⎟
⎜ FPschn ⎟ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ 0 · · · JPschn ⎟ ⎜ ΔΦPschn ⎟
⎜ FPass1 ⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ JPass1 · · · 0 ⎟ ⎜ ΔΦPass1 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ .
.. ⎟ ⎜ 0 0 . . . [J ] ⎟ ⎜ ⎜ .. ⎟

⎜ ⎟ ⎜ . . . Padc ⎟ .
⎟ ⎜ ⎟
. . .
⎝ FPassr ⎠ ⎜
⎝ 0 · · · JPassr ⎠ ⎝ ΔΦPassr ⎠
FDC [JdcS ] [JdcPs ] [JdcPa ] [Jdc ] ΔΦDC
(2)

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
2892 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 11, NO. 4, OCTOBER 2020

Fig. 4. Control loop for the reactive power control at stator’s terminals.

to the switching power loss at rated values. Ir and Ig are the


Fig. 2. Schematic diagram of a DFIG-based wind generator. actual terminal currents of the RSC and GSC, respectively, and
Inom is for the converters’ nominal currents.
The expressions for the active and reactive power injections
at nodes r and 0r, with Vr = Vr ∠θr and Vg = Vg ∠θg , are
Pr = Vr2 Gr − kmr Vr Edc [Gr cos (θr − φr )
+ Br sin (θr − φr )] (3)
Qr = − Vr2 Br − kmr Vr Edc [Gr sin (θr − φr )

Fig. 3. Equivalent circuit of the B2B converter. − Br cos (θr − φr )] (4)


P0r = k 2 2 2
mr Edc Gr − kmr Vr Edc [Gr cos (φr − θr )
converter, which seamlessly combines with the rest of the com- + Br sin (φr − θr )] + Pswr (5)
ponents comprising the DFIG-type wind generator model with
Q0r = − k 2 mr 2 Edc
2
Br − kmr Vr Edc [Gr sin (φr − θr )
the schematic diagram shown in Fig. 2.
− Br cos (φr − θr )] + Qeqr (6)
A. The New B2B Converter for DFIG-Based Wind Generators
where Pswr , Qeqr and the terminal current Ir are computed as,
The power electronic circuit employed in a DFIG is made
up of two VSCs connected in a B2B arrangement sharing a Pswr = Edc
2
G0r (Ir /Inom )2 (7)
capacitor on their DC side, as illustrated schematically in Fig. 2. Qeqr = −k22 mr 2 Edc
2
Beqr (8)
The DC voltage is supported and stabilized by the DC capacitor,
whereas the VAR generation/absorption process is attained by Ir =
the PWM control via the manipulation of the voltage and current
waveforms inside the VSCs. (G2r + Br2 ) (Vr2 + k 2 mr 2 Edc
2 − 2km V E cos (θ − φ ))
r r dc r r

The DC-to-AC voltage conversion for a two-level, three-phase (9)


VSC is V = kmEdc ejφ , where the voltage V is the line-to-line A similar set of equations to (3)–(9) may be obtained for the
AC voltage at the network’s reference frame. Edc is the DC powers injected at the GSC terminals by simply replacing the
link voltage, k = (3/8), m is the amplitude modulation ratio, subscripts r by g, where node g would correspond to a generic
and φ is the phase-shifting angle of the voltage V with respect node k of the system, therefore Vk = Vg and θk = θg . Also,
to Edc . To capture these features of the VSC operation, at notice that in the previous equations and their state variables are
the fundamental frequency, the rotor-side converter (RSC) and expressed in the network’s reference frame. The rotor voltages
the grid-side converter (GSC), may be represented each as an in the dq reference frame, vdr and vqr , may be then obtained by
ideal tap-changing transformer with complex taps, coupled to using the standard transformation,
an impedance, an equivalent variable shunt susceptance and a
shunt resistor on its DC side, as shown in Fig. 3. Vr = vdr 2 + v 2 , θ = arctan (V
qr r r,imag /Vr,real ) (10)
The tap magnitudes of the ideal tap-changing transformers



bear a resemblance to the amplitude modulation ratios of the Vr,real cos δr sin δr vqr
= ,
RSC and GSC, mr and mg , respectively. The phase-shifting Vr,imag sin δr − cos δr vdr
angles φr and φg are the phase angles of voltages V1r and
δr = θe − (ωm /ωs ) θk (11)
V1g , respectively. The series reactances Xr and Xg represent the
interface magnetics. The series resistor Rr and Rg are associated where δr represents the difference between the angular displace-
with the conduction losses, of RSC and GSC, respectively, which ment θe of the synchronous rotating reference frame and the
are proportional to the AC terminal current squared. These re- induction machine’s rotor angle [26].
actances and resistances also account for the AC filtering equip- Rotor-side converter controllers: the controllers of the RSC
ment between the VSCs and the rotor/stator circuits. The con- are shown in Fig. 4 and Fig. 5. The control loop that exerts
ductances Gswr = G0 (Ir /Inom )2 and Gswg = G0 (Ig /Inom )2 voltage control over vdr is intended for keeping a reference value
enable the calculation of the switching power losses of the of reactive power in the stator as Qgs,ref = Pnom tan(ϕ), where
converters, where G0 is a constant conductance corresponding ϕ stands for the fixed power factor angle at which the stator is

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
CASTRO AND ACHA: ON THE DYNAMIC MODELING OF MARINE VSC-HVDC POWER GRIDS INCLUDING OFFSHORE WIND FARMS 2893

Fig. 5. Control loop for maximum power extraction.

Fig. 7. (a) Blade pitch angle controller; (b) MPT characteristic.

wind speed Vw , the area swept by the rotor blades A, its aerody-
Fig. 6. (a) DC voltage controller of the B2B converter, (b) AC voltage
namic power coefficient Cp , the air density ρ, the tip-speed ratio
controller of the GSC. λ and the blade pitch angle β [27]. Moreover, there is a power
limitation strategy to prevent mechanical stress in the rotor shaft
when the wind speed surpasses the rated speed. The mechanical
required to operate [16], [17], as shown in Fig. 4. On the other power is reduced using a controller, which acts upon the blade
hand, the control loop shown in Fig. 5 drives the DFIG to operate pitch angle, β, as shown in Fig. 7(a). One way to design the pitch
at the maximum power extraction point (MPT). This controller angle controller calls for the monitoring of the electrical power
regulates the rotor voltage vqr in accordance with changes in the output of the DFIG, Pe = Pgs − Pg , to bring it back within
actual rotor speed ω m . limits by changing β. On the other hand, a power optimization
Grid-side converter controllers and DC link dynamics: when strategy is implemented in the DFIG which focuses on driving
changes occur in the DC link, the capacitor undergoes charg- the wind generator to operate at its optimum rotational speed
ing/discharging stages, which impacts on its DC voltage. Apply- and torque, Tref = Pm,opt /ωm,opt , as shown in Fig. 7(b).
ing Kirchhoff’s law to the DC link shown in Fig. 2, the following Wound-rotor induction generator model: The voltage equa-
relationship holds: ic = idcr − idcg . By equating this expression tions that govern the WRIG’s dynamic behavior are obtained
and that of the capacitor’s current, yields: by applying Kirchhoff’s and Faraday’s laws as well as Park’s
transformation and its third-order dynamic model is expressed
dEDC idcr − idcg
= (12) as [26],
dt Cdc
de d (ωs − ωm ) Lr  ω s Lm
where the current idcr is computed as idcr = −P0r /Edc , with T0 = ωs Lm iqr + eq − vqr (14)
dt Rr Rr
P0r taken from (5). Substituting this expression into (12), the
DC link voltage dynamics in the B2B converter are: de q (ωs − ωm ) Lr  ω s Lm
T0 = −ωs Lm idr − ed + vdr (15)

dt Rr Rr
dEDC 1 P0r
= − − idcg (13) dωm Te − Tm
dt CDC EDC = (16)
dt 2H
As it is a standard practice, the GSC is responsible for con-
trolling the voltage at the DC link regardless of the amount where Rr , Lr and Lm are the rotor resistance, rotor inductance
and direction of the rotor active power Pgr . The DC voltage and magnetizing inductance, e´ is the per-unit e.m.f behind the
control is carried out by controlling the DC current entering the transient reactance, ωs is the electrical rotating sped of the stator,
GSC, idcg , as shown in Fig. 6(a), where Edc and Edcref are the ωm is the mechanical speed of the generator rotor, T’0 is the
actual and reference voltages, respectively. On the other hand, transient open-circuit time constant [s], Tm is the mechanical
the GSC also performs AC voltage regulation, therefore only torque, Te is the electromagnetic torque, H is the total inertia
one basic control loop for the modulation ratio of the GSC, mg , of the rotating mass, v is voltage, i is current, the subscripts d y
is required, as shown in Fig. 6(b), where Vk and Vkref are the q stand for the direct and quadrature axes, respectively, and the
(0) subscripts s y r represent quantities corresponding to the stator
actual and reference voltages, respectively, and mg is the initial,
and rotor, respectively.
steady-state value of the modulation ratio.
IV. VSC-CONNECTED DFIG-BASED WIND FARMS
B. Modeling of the Wind Turbine and the WRIG
For completeness, the modeling of its wind turbine and its A. Steady-State Operating Conditions of Each DFIG
controls and the wound rotor induction generator, are briefly The steady-state model of the DFIG may be directly derived
discussed in this section. from the algebraic power equations and differential equations
Wind turbine and its basic controls: the conversion of the (with the dynamic terms set to zero) of the WRIG (14)–(16)
kinetic energy from the wind into mechanical power is made which combine well with those of the B2B converter (2)–(7),
by the wind turbine’s rotor and may be computed by Pm = keeping in mind that the RSC controllers ensure that the DFIG
1/2ρCp (c1 . . . c9 , λ, β)AVw3 , which is mainly a function of the operates at Qgsref and Tref . If the DFIG is connected to node

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
2894 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 11, NO. 4, OCTOBER 2020


k of the network, the iterative Newton-Raphson method may be Fe q = eq(t−Δt) + 0.5Δt ė q(t−Δt) − e q(t) − 0.5Δt ė q(t)
used to solve the set of nonlinear equations (17)–(27), whose (35)
solution may be used as the steady-state operating conditions
necessary for the ensuing dynamic simulations. Fωm = ωm(t−Δt) +0.5Δt ω̇m(t−Δt) − ωm(t) − 0.5Δt ω̇m(t)
(36)
ΔPk = Pgs − Pg − Pdk − Pkcal (17)
Fβaux = βaux (t−Δt) + 0.5Δt β̇aux(t−Δt)
ΔQk = Qgs − Qg − Qdk − Qcal
k (18)  
− βaux (t) − 0.5Δt β̇aux(t) (37)
Δχ1 = ωs Lm iqr + (ωs − ωm ) Lr Rr−1 eq − ωs Lm Rr−1 vqr  
(19) Fβ = β(t−Δt) + 0.5Δt β̇(t−Δt) − β(t) − 0.5Δt β̇(t) (38)
Δχ2 = −ωs Lm idr − (ωs − ωm ) Lr Rr−1 ed + ωs Lm Rr−1 vdr
Fidr,aux = idr,aux (t−Δt) + 0.5Δt i̇dr,aux(t−Δt)
(20)
 
ΔT = Tref − Te (21) − idr,aux (t) − 0.5Δt i̇dr,aux(t) (39)

ΔQgs = Qgsref − Qgs (22) Fvdr,aux = vdr,aux (t−Δt) + 0.5Δt v̇dr,aux(t−Δt)


ΔPr = Pgr − Pr (23)  
− vdr,aux (t) − 0.5Δt v̇dr,aux(t) (40)
ΔQr = Qgr − Qr (24)
Fvqr,aux = vqr,aux (t−Δt) + 0.5Δt v̇qr,aux(t−Δt)
ΔQ0r = Q0r (25)
 
ΔQ0g = Q0g (26) − vqr,aux (t) − 0.5Δt v̇qr,aux(t) (41)
ΔP0 = P0r + P0g (27) FEdc = Edc(t−Δt) + 0.5Δt Ėdc(t−Δt)
 
where Pdk and Qdk represent the active and reactive powers − Edc(t) − 0.5Δt Ėdc(t) (42)
drawn by a load at bus k, respectively, and Pkcal and Qcal k are
the active and reactive power terms that link the node k with the Fidcaux = idcaux(t−Δt) + 0.5Δt i̇dcaux(t−Δt)
rest of the network. The solution of (17)–(27), using Newton’s
− idcaux(t) − 0.5Δt i̇dcaux(t) (43)
algorithm, requires the formation of the Jacobian matrix. The
linearization of this set of equations is ΔF = −J Δz, where J Fdmag = dmg(t−Δt) + 0.5Δt dṁg(t−Δt)
represents a matrix of first order partial derivatives, ΔF = [ΔPk
ΔQk Δχ1 Δχ2 ΔT ΔQgs ΔPr ΔQr ΔQ0r ΔQ0g ΔP0 ]T and − dmg(t) − 0.5Δt dṁg(t) (44)
 
Δz = [Δθk Δmg Δvdr Δvqr Δed Δeq Δφr Δmr ΔBeqr ΔBeqg Δφg ].
Notice that the vector Δz includes neither Edc nor Vk because The linearized form of the dynamic DFIG model, connected
both terms are constant during the steady-state period. to a generic node k of the wind farm distribution network, is

(i,t) (i,t) (i,t)


ΔFDF IG = −JDF IG ΔzDF IG (45)
B. DFIG Model for RMS-Type Dynamic Simulations
Once the initial conditions have been determined, the dynamic ΔF (i,t) = [ΔPk ΔQk ΔPr ΔQr ΔQ0r ΔP0 Fe d Fe q . . .
simulations may be carried out by numerically solving, step- Fωm Fβaux Fβ Fidr,aux  Fvdr,aux Fvqr,aux
by-step, all the power mismatch equations and the discretized FEdc . . . Fidcaux Fdmg
differential-algebraic equations, also expressed as mismatch (46)
equations (using the implicit trapezoidal method), emerging Δz (i,t) = [Δθk ΔVk Δφr Δmr ΔBeqr Δφg Δe d Δe q . . .
from the dynamic controllers of each DFIG, Δωm Δβaux Δβ Δidr,aux Δvdr,aux . . .
Δvqr,aux ΔEdc Δidcaux Δdmg ]T
ΔPk = Pgs − Pg − Pdk − Pkcal (28)
(47)
ΔQk = Qgs − Qg − Qdk − Qcal
k (29)
ΔPr = Pgr − Pr (30) C. Dynamic Model of VSC-Connected DFIG-Based
Wind Farms
ΔQr = Qgr − Qr (31)
Fig. 8 shows, in schematic form, a power converter, VSCW F ,
ΔQ0r = Q0r (32) linking a wind farm and a generic DC network. This power
ΔP0 = Edc idcg − P0g (33) converter facilitates the free transit of the power generated by
the wind generators, continuously pursuing the balance between
F e d = ed(t−Δt) + 0.5Δt ė d(t−Δt) − e d(t) − 0.5Δt ė d(t) the power produced by the wind farm and the power injected into
(34) the DC network. However, this power balance cannot be attained

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
CASTRO AND ACHA: ON THE DYNAMIC MODELING OF MARINE VSC-HVDC POWER GRIDS INCLUDING OFFSHORE WIND FARMS 2895

Fig. 9. (a) DC current controller and (b) modulation ratio controller of the
converter VSCWF.
Fig. 8. VSC connected DFIG-based wind farm coupled to a DC grid.

instantaneously, causing the wind farm’s AC power network to the VSCWF , Idc , aimed at keeping the frequency of its coupled
experience frequency deviations. wind farm at its nominal value, and the other control loop for
From the power system standpoint, the VSCW F must provide exerting AC voltage control using the modulation ratio, as shown
the angular reference for all the nodal voltage phase angles in Figs. 9(a) and 9(b), respectively.
of the wind farm AC power network, i.e., φwf = 0, during In connection with Fig. 8, the overall mismatch equations of
steady-state. For the dynamic operating regime, the difference the voltage source converter VSCW F , including those resulting
between the DC power entering the inverter, Edc Idc , and the from its ancillary components, are given by (50)–(59). The
power flowing out of the converter towards its AC terminals, P0v , discretized differential equations of the voltage dynamics of Cdc
will yield frequency deviations in the wind farm. Indeed, P0v and current dynamics of Ldc of VSCW F are given in (50)–(51),
corresponds to the total power generated by the wind generators respectively. Equations (52)–(55) are obtained by discretizing
minus the power losses produced by the wind farm AC power the differential equations emerging from the control loops shown
network. The angular frequency ωwf and the angle φwf of the in Fig. 9 and from (48), respectively. Finally, equations (56)–(59)
converter may be dynamically computed as stand for the active and reactive power mismatch equations at
nodes v and k, respectively.
dωwf ω0
(a) = (Edc Idc − P0v ) , FEdc = Edc(t−Δt) + 0.5ΔtĖdc(t−Δt)
dt 2Hvsc
 
dφwf − Edc( t) − 0.5ΔtĖdc(t) (50)
(b) = ωwf − ω0 (48)
dt  
where ω0 = 2πfnom and fnom is the nominal frequency of FIx = Ix(t−Δt) + 0.5ΔtI˙x(t−Δt) − Ix( t) − 0.5ΔtI˙x(t)
the wind farm. Notice that the angle φwf and the angular (51)
speed ωwf are also the reference signals for the wind farm
AC power grid. The equivalent inertia constant of the converter Fdma = dma(t−Δt) + 0.5Δtdṁa(t−Δt)
Hvsc [s] may be estimated from the electrostatic and electro-
− dma(t) − 0.5Δtdṁa(t) (52)
magnetic energies stored in the capacitor, WC = 1/2; Cdc E02 ,
and inductor, WL = 1/2; Ldc I02 , coupled to its DC bus. This FIdcaux = Idcaux(t−Δt) + 0.5ΔtI˙dcaux(t−Δt)
fictitious inertia constant of the VSCW F may be calculated  
by Hvsc = (WC + WL )/Snom [18], where E0 and I0 are the − Idcaux(t) − 0.5ΔtI˙dcaux(t) (53)
nominal voltage and current of the capacitor and inductor, re-
spectively, whereas Snom is the rated apparent power of the Fφwf = φwf (t−Δt) + 0.5Δtφ̇wf (t−Δt)
converter. This takes an approximate value of Hvsc ≈ 5 ms,  
which is much smaller than the inertia featured by conventional − φwf (t) − 0.5Δtφ̇wf (t) (54)
rotating machines (∼5 s). The power flowing out of the converter
Fωwf = ωwf (t−Δt) + 0.5Δtω̇wf (t−Δt)
towards its AC terminals, is computed as,
 
P0v = k 2 m2a Edc
2
G − kma Vv Edc [G cos (φwf − θv ) − ωwf ( t) − 0.5Δtω̇wf (t) (55)

+ B sin (φwf − θv )] + Edc


2
G0 Iv2 /Inom
2
(49) ΔPv = − Pv − Pdv − Pvltc (56)
where = (G + B )
Iv2 2 2
(Vv2
+k 2
− 2kma Vv Edc
2 2
ma Edc ΔQv = − Qv − Qdv − Qvltc (57)
cos(θv − φwf )), Inom is the nominal current of the power
converter, Y = G + jB accounts for both the conduction losses ΔPk = − Pkltc − Pdk − Pkcal (58)
and interface
 magnetics of the converter and the phase reactor, ΔQk = − Qkltc − Qdk − Qcal (59)
k
k = 3/8, Vv = Vv ∠ θv , and G0 is related to the switching
losses of the VSCW F . For unified dynamic simulations of VSC connected DFIG-type
The control system of a VSC interfacing a wind farm with wind farms, the previous set of equations, FVSCwf : (50)–(59),
a DC network may be realized by using two control loops to must be solved, at every integration step Δt, together with
conform to practical converter stations, which offer two degrees those corresponding to each DFIG included in the wind farm,
of freedom. One control loop for regulating the DC current of FDFIG : (28)–(44), along with those resulting from the power

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
2896 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 11, NO. 4, OCTOBER 2020

TABLE II power mismatch equations of the high-voltage side of the OLTC


CONVERTER VSCWF AND ASSOCIATED VARIABLES
(see Fig. 1) with respect to their corresponding VSC’s variables
(shown in Table I), for the three types of converters (VSCSlack ,
VSCP sch and VSCP ass ). Similarly, the derivatives of the active
and reactive power mismatch equations of the high-voltage side
of the OLTC with respect to their corresponding AC voltage
mismatch equations of the wind farm AC network, ΔPQwf . magnitudes and phase angles are suitably accommodated in
Hence, if the DFIG-type wind farm contains n wind generators, JMTm , JMTn and JMTr , for the slack, power-scheduled and
its dynamical model will have the following numerical structure: passive converters, respectively.
where JVSC wf accommodates the partial derivatives of the The terms FWF γ, JWF γ and ΦWF γ, which are explicitly
mismatch equations of VSCW F , (50)–(59), with respect to its given in (63), comprise nwf individual terms of the form of
state and control variables ΔzVSC wf , summarized in Table II. (60)–(61). Matrices JWFdc accommodate the terms arising
The coupling between the VSC and the wind farm AC grid is from deriving (50)–(51), of each VSCW F , with respect to all
node k, as shown in Fig. 8; hence, JV wf contains the deriva- nodal voltages of the DC network. Similarly, matrices JdcWF
tives of (58)–(59) with respect to all nodal phase angles and comprise the derivative terms of DC network’s mismatch equa-
voltage magnitudes that link to node k, through a low-voltage tions with respect to the DC voltages of each VSCW F .
AC transmission line or cable. Similarly, Jwf V comprises the ⎛ ⎞ ⎛ ⎞
derivatives of active and reactive power mismatches of the wind FWF1 JWF1 0
⎜ .. ⎟ ⎜ .. ⎟
farm’s nodes with respect to θk and Vk . The Jacobian matrix Jac FWFγ = ⎝ . ⎠, JWFγ = ⎝ . ⎠,
contains the partial derivatives of the active and reactive power FWFnwf 0 JWFnwf
mismatches with respect to the phase angles and the voltage ⎛ ⎞
ΔΦWF1
magnitudes and ΔθV is the vector of corrections of the phase ⎜ .. ⎟
angles and the voltage magnitudes of the wind farm’s AC power ΔΦWFγ =⎝ . ⎠,
network. Jac1 , . . . , Jacn are vectors of suitable orders which ΔΦWFnwf
accommodate the derivatives of the active and reactive power ⎛ ⎞T ⎛ ⎞
JdcWF1 JWFdc1
mismatches with respect to the state and control variables of each ⎜ .. ⎟ ⎜ .. ⎟
DFIG contained in the wind farm. The derivatives of the DFIG’s JdcWF =⎝ . ⎠ , JWFdc = ⎝ . ⎠
mismatch equations with respect to the phase angles and voltage JdcWFnwf JWFdcnwf
magnitudes of the wind farm AC power network are suitably (63)
located in vectors J1ac , . . . , Jnac . Notice that FDFIGj , JDFIGj
and ΔzDFIGj , with j = 1, …n, take the form of (45)–(47). It should be noted that (62)–(63) can be suitably modified to
In a compact manner, the dyanamic model of a VSC- incorporate alternative control strategies, such as the Edc − P
connected DFIG-based wind farm (60) shown at the bottom of control droop strategy for the VSC-HVDC grid. This would
this page, is as follows: require a suitable reformulation of the algebraic differential
equations emerging from the control loops of the VSCs shown
(i,t) (i,t) (i,t)
FWF = −JWF ΔΦWF (61) in the Appendix. It is worthwhile remarking that an efficient dy-
namic simulation heavily hinges on selecting a suitable starting
V. DYNAMIC MODELING FRAMEWORK FOR MULTI-TERMINAL equilibrium point. This is obtained by solving the steady-state
VSC-HVDC CONNECTED DFIG-BASED WIND FARMS power flow equations of the VSC-based transmission grid with
their coupled AC systems as in [18], [19], including those of the
The dynamic model of the multi-terminal VSC-HVDC sys- newly developed VSC-connected offshore wind farms (17)–(27)
tem, given by JMT−HVDC in (2), comprising m VSCSlack and (56)–(59).
converters, n VSCP sch converters and r VSCP ass converters,
with α = m + n + r, is extended to comprise nwf DFIG-type
wind farms. The newly expanded framework is given by (62). VI. STUDY CASES
Expression (62) shown at the bottom of the next page, in- The numerical implementation of the developed formulation
cludes, for completeness, the contribution of the conventional was coded in MATLAB and the simulations were executed
AC power networks (FACα , JACα , ΦACα ), which are coupled using a PC Intel Core i7, 3.6-GHz, 8 GB RAM, Win 7 64-bit
to the DC network through VSC units. Matrices, JACm , JACn OS. The differential equations related to the VSC and DFIG
and JACr contain the derivative terms of the active and reactive controls are discretized with the implicit trapezoidal method,

⎛ ⎞(i,t) ⎛ ⎞(i,t) ⎛ ⎞(i,t)


FVSCwf JVSCwf JV wf 0 ΔzVSCwf
⎜ ΔPQwf ⎟ ⎜ Jwf V Jac Jac1 ··· Jacn ⎟ ⎜ ΔθV ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ FDFIG1 ⎟ ⎜ J1ac JDFIG1 0 0 ⎟ ⎜ ΔzDFIG1 ⎟
⎜ ⎟ = −⎜ ⎟ ⎜ ⎟ (60)
⎜ .. ⎟ ⎜ .. .. ⎟ ⎜ .. ⎟
⎝ . ⎠ ⎝ 0 . 0 . 0 ⎠ ⎝ . ⎠
FDFIGn Jnac 0 0 JDFIGn ΔzDFIGn

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
CASTRO AND ACHA: ON THE DYNAMIC MODELING OF MARINE VSC-HVDC POWER GRIDS INCLUDING OFFSHORE WIND FARMS 2897

wind farm I increases from 8 m/s to 10 m/s with a rate of change


of ΔVw /Δ = 1 m · s−1 /s.
The steady-state conditions of selected variables for the VSC-
based transmission grid with two VSC-connected wind farms are
reported in Table III. This table also shows the corresponding re-
sults obtained for the switching-based AC/DC power grid model
implemented in Simulink, after 50 s of simulation. It is observed
that relatively small differences exist between both approaches,
i.e., average errors for the DC power flows and variables are infe-
rior to 3%. Similarly, Fig. 11 reports the validation of results for
the dynamic operating regime of the VSC-based HVDC power
grid, with both models being executed for 50 s of simulation.
Notice that the dynamic response furnished by the introduced
approach concurs very well with that of the highly-detailed
EMT-type solution. Table IV reports the computing times of
Fig. 10. Six-terminal HVDC link with two VSC-connected wind farms. both fundamentally different approaches where the superior
computational efficiency featured by the developed method is
evident.
using an integration time step of 1 ms. In turn, the numerical Note that the model initialization (steady-state conditions) for
solution of the differential-algebraic equations and the nodal the EMT-type simulation takes various hours because of the very
active and reactive power balance equations for each of the different dynamics (time constants) of the generation units and
AC and DC nodes is obtained by adopting a unified method, VSC stations. That is, the power generators contained in the
signifying that the nonlinear equations (61)–(62) representing AC grids were realistically modeled, as opposed to representing
the marine HVDC-connected wind farms are simultaneously them as idealized, controlled voltage sources. Certainly, once the
solved at every time step. steady-state conditions have been obtained, these can be saved
The AC/DC network shown in Fig. 10 is used to illustrate as a snapshot and used for starting new dynamic simulations,
the validity and prowess of the new modeling framework. This thus avoiding the recalculation of the steady-state. However, this
is an HVDC grid made up by six VSC stations, two of which recalculation would be needed if the AC/DC grid model were up-
connect to identical 200-MW wind farms. This power grid model dated, a situation that occurs when power system analysts carry
was also implemented in the Simscape Electrical software of out various dynamic simulations at the planning and operation
Simulink, for comparison purposes. stages. These aspects justify per se the new RMS-type dynamic
VSC 1 is taken to be the slack converter, VSCSlack , which modeling framework for HVDC systems with VSC-connected
exerts DC voltage control, VSC 2 feeds into a passive network wind farms. It should be kept in mind that Transmission System
(PN) having a 100 MW load whereas VSC 3 and VSC 5 are Operators (TSO) often resort to using supercomputers to reduce
power-scheduled converters, VSCP sch , drawing each Psch = the computational times. However, it is expected that the ratio
125 MW from the DC network. The DFIG-type wind farms, of computing times between both approaches is similar given
WF I and WF II, are connected to the DC network through the two fundamentally different approaches under comparison.
the stations VSC 4 and VSC 6, respectively. With no loss of Concerning the dynamics involved in this study case, the
generality, each wind farm is represented by three equivalent sudden increase of Psch in VSC 3 and, hence, in P3, as seen
wind generators. On the other hand, the three AC networks, in Fig. 11(a), causes a transient discharge of the energy stored
AC 1, AC 2 and AC 3, which are coupled to VSC 1, VSC 3 and in the DC capacitors, causing nodal voltage drops in the DC
VSC 5, respectively, are represented by the Thevenin equivalents network, as shown in Fig. 11(b). As soon as the DC voltage
shown in Fig. 10. The transmission line impedances of these AC controller of VSC 1 (which acts as VSCSlack ) takes over, this
grids, including those of the wind farms, are z = j0.06 p.u and converter starts injecting power into the DC network, as seen in
the DC cable resistances are shown in the figure. The rest of the P1 of Fig. 11(a), resulting in an improved DC voltage. Notice
parameters are given in the Appendix. that this perturbation produces a negligible impact on the other
Study case I - validation of the new approach: Two events are VSC connected networks, i.e., PN, WF I, AC 3 and WF II. On
simulated to assess the power grid dynamic response: (i) at t = the other hand, the wind speeds of WG 1, 2 and 3 comprising
0.5 s, the scheduled power, Psch , of VSC 3 abruptly changes WF I, start increasing after t = 10 s, reaching each a mechanical
from 125 MW to 200 MW; (ii) at t = 3 s, the wind speed, Vw , of torque of Tm = −0.458 p.u. and a wind power generation of

⎛ ⎞(i,t) ⎛ ⎞(i,t) ⎛ ⎞(i,t)


FACα JACα JACm JACn JACr 0 0 ΔΦACα
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ FSlackm ⎟ ⎜ JMTm 0 ⎟ ⎜ ΔΦSlackm ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ FPschn ⎟ ⎜ JMTn 0 ⎟ ⎜ ΔΦPschn ⎟
⎜ ⎟ = −⎜ JMT−HVDC ⎟ ⎜ ⎟ (62)
⎜ FPassr ⎟ ⎜ JMTr 0 ⎟ ⎜ ΔΦPassr ⎟
⎝ ⎠ ⎜ ⎟ ⎝ ⎠
FDC ⎝ 0 JdcWF ⎠ ΔΦDC
FWFγ 0 0 0 0 JWFdc JWFγ ΔΦWFγ

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
2898 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 11, NO. 4, OCTOBER 2020

TABLE III
STEADY-STATE CONDITIONS OF THE SIX-TERMINAL VSC-BASED HVDC NETWORK

Fig. 11. Validation of the proposed approach - Study case I. (a) Power entering the VSCs; (b) DC grid voltages; (c) Modulation ratios; (d) Mechanical torque;
(e) Output power by the wind farms.

TABLE IV AC 1, as observed in Fig. 11(a). On the other hand, the dynamic


COMPUTING TIMES
performances of the modulation ratios of the six VSCs, shown
in Fig. 11(c), are dynamically adjusted to comply with the AC
voltage control command. Note that the VSC modulation ratios
experience greater transient oscillations for the abrupt change in
Psch of VSC 3 occurs than when WF I varies its output power
generation. Indeed, the variations of Vw in WF I, only impact
the DC network and the AC power grid coupled to the VSCSlack
Pg = 113.49 MW, at the new steady-state when Vw = 10 m/s type station, AC 1, with the other power grids experiencing
and t = 40 s approximately, as observed from Figs. 11(d)–11(e), minor variations. This is the hallmark of an HVDC system,
respectively. This results in additional power entering the DC which blocks the electrical interactions between AC power
network thus causing minor voltage swells. The additional networks. However, it may be observed that when power changes
power in the DC network is counteracted by VSC 1, which occur at any of the VSC connected networks, the AC power
reduces its power P1 , assisting in the power balance of network grids coupled to VSCSlack type converters do experience power

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
CASTRO AND ACHA: ON THE DYNAMIC MODELING OF MARINE VSC-HVDC POWER GRIDS INCLUDING OFFSHORE WIND FARMS 2899

Fig. 13. Dynamic behavior of the HVDC system – Study case III.

Fig. 12. Dynamic behavior of the HVDC system – Study case II.

deviations. This situation is expected to occur in the context of


multi-terminal HVDC links with VSC-connected DFIG-based
wind farms. To shed additional light on this point, the Study Case
II is carried out.
Fig. 14. Control loops for converters: (a) VSCSlack ; (b) VSCP ass ; (c)
Study case II: Considering that VSC 1 and VSC 5 are slack VSCP sch ; (d) Voltage control for the three VSCs
converters, at t = 0.5 s: (i) the wind speed of wind farm I
increases from 8 m/s to 10 m/s; (ii) the wind speed of wind
during the disturbance. As expected, this internal contingency
farm II decreases from 12 m/s to 8 m/s. Both speed changes
of the HVDC system imposes little impact on the AC powers,
are characterized by ΔVw /Δt = 4 m · s−1 /s; as expected, these
including those of the wind farms, P4 and P6 , as appreciated
are directly reflected on the mechanical speed of the wind
from Fig. 13(b). However, this disturbance does cause power
generators, as illustrated in Fig. 12(a). Furthermore, the power
flow rearrangements, accompanied by voltage fluctuations, in
generated by WF I increases from 60.21 MW to 117.3 MW
the HVDC grid, as shown in Fig. 13(c). In turn, these are
whereas that of WF II reduces from 188.26 MW to 60.02 MW,
efficiently mitigated by VSC 1 and 5 which fittingly control
as shown in Fig. 12(b), a fact that is reflected on powers, P4 and
their DC bus voltages. In contrast, Fig. 13(d) shows that the
P6 , of VSC 4 and VSC 6, respectively, as observed in Fig. 12(c).
modulation ratios barely change during the event.
Recalling that VSC 1 and VSC 5 are tasked with DC voltage
Admittedly, this kind of bottleneck scenarios may be typical
control, as opposed to DC power control, both converters share
in the operation of HVDC grids with wind power generation,
the responsibility of balancing the powers in the AC/DC system.
thus the developed dynamic modelling framework may become
Hence, their coupled networks, AC 1 and AC 3, must supply
a very useful tool for power system operators.
or absorb the necessary power to attain this balance, therefore
frequency deviations appear in both AC systems, as shown in
VII. CONCLUSION
Fig. 12(d), settling down to 49.92 Hz and 49.98 Hz, respectively.
The wind power variations also cause temporary frequency devi- This paper has introduced an advanced modeling frame-
ations in the AC grids of the DFIG-based wind farms. However, work for the efficient and reliable steady-state and dynamic
converters VSCW F I and VSCW F II fittingly control the current assessments of multi-terminal VSC-HVDC systems with VSC-
injection to the DC grid thus effectively regulating the frequency connected DFIG-based wind farms. From the modeling stand-
of both wind farms, as inferred from Fig. 12(d). The dynamic point, the new frame-of-reference is comprehensive and highly
performances of the DC voltages and modulation ratios of the flexible; it can accommodate any number of VSC stations with
VSCs are given in Figs. 12(e) and 12(f), respectively, noticing different control strategies to conform to their pairing AC grid
that both variables experience very small variations in this study requirements. From the literature survey and the validation test
case. carried out in the paper, it is concluded that heretofore the
Study case III: The DC transmission line that connects buses introduced modeling framework is more flexible and numer-
3-4 trips at t = 0.5 s. This event provokes the meshed DC grid ically efficient than alternative formulations based on EMT
to turn into a radial one. As illustrated in Fig. 13(a), this DC line models.
carried 116 MW at steady-state, but its disconnection prevents Both the steady state and the dynamic formulations are solved
part of the wind power generation of WF I from being transmitted using the Newton-Raphson method; the latter is combined with
to the passive network fed by VSC 2. In this condition, the only the implicit trapezoidal rule. This applies to conventional power
path available to evacuate the wind generation by WF I is the DC plants and their controls, the VSC-connected DFIG-based wind
line 4–5, whose power flow changes from −60 MW to 59 MW farms and the VSC controls of the multi-terminal HVDC system.

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.
2900 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 11, NO. 4, OCTOBER 2020

The rotor and stator circuits of the developed RMS-type DFIG [5] H. J. Bahirat and B.A. Mork, “Operation of DC series-parallel connected
model seamlessly combine with its HVDC converter model. offshore wind farm,” IEEE Trans. Sust. Energy, vol. 10, no. 2, pp. 596–603,
Apr. 2019.
Moreover, the explicit representation of the DFIG’s back-to- [6] P. Hou, W. Hu, M. Soltani, C. Cheng, B. Zhang, and Z. Chen, “Offshore
back converter model brings about unrivalled modeling flexibil- wind farm layout design considering optimized power dispatch strategy,”
ity. This is further accentuated when the DFIG-based wind farms IEEE Trans. Sustain. Energy, vol. 8, no. 2, pp. 638–647, Apr. 2017.
[7] O. Dahmani, S. Bourguet, M. Machmoum, P. Guerin, P. Rhein, and L.
are placed within the context of multi-terminal VSC-HVDC Josse, “Optimization and reliability evaluation of an offshore wind farm,”
systems. IEEE Trans. Sustain. Energy, vol. 8, no. 2, pp. 542–550, Apr. 2017.
Ample use of the modeling framework has shown to be a very [8] Y. Guo, H. Gao, Q. Wu, H. Zhao, J. Ostergaard, and M. Shahidehpour,
“Enhanced voltage control of VSC-HVDC-connected offshore wind farms
reliable computational tool. To demonstrate this, a six-terminal based on model predictive control,” IEEE Trans. Sustain. Energy, vol. 9,
VSC-HVDC link with two VSC-connected DFIG-based wind no. 1, pp. 474–487, Jan. 2018.
farms was used. It is shown that the dynamic interactions be- [9] L. P. Kunjumuhammed, B. C. Pal, C. Oates, and K. J. Dyke, “Electrical
oscillations in wind farm systems: Analysis and insight based on detailed
tween the interconnected networks have been well captured by modeling,” IEEE Trans. Sustain. Energy, vol. 7, no. 1, pp. 51–62, Jan. 2016.
the modeling and numerical solution approach presented in the [10] U. Karaagac, S.O. Faried, J. Mahseredjian, and A. Edris, “Coordinated
paper. As a general conclusion, it may be surmised that the control of wind energy conversion systems for mitigating subsynchronous
interaction in DFIG-based wind farms,” IEEE Trans. Sustain. Energy,
potential usefulness of the modeling, techniques and methods vol. 5, no. 5, pp. 2440–2449, Sep. 2014.
derived in the paper would be of paramount importance to power [11] H. J. Bahirat and B. A. Mork, “Operation of DC series-parallel connected
system operators and analysts when dealing with this type of offshore wind farm,” IEEE Trans. Sustain. Energy, vol. 10, no. 2, pp. 596–
603, Apr. 2019.
future HVDC power networks. [12] S. Cole, J. Beerten, and R. Belmans, “Generalized dynamic VSC MTDC
model for power system stability studies,” IEEE Trans. Power Syst., vol. 25,
no. 3, pp. 1655–1662, Aug. 2010.
APPENDIX [13] G. O. Kalcon, G. P. Adam, O. Anaya-Lara, S. Lo, and K. Uhlen, “Small
signal stability analysis of multi-terminal VSC-based DC transmission
– Parameters of each VSC on a 100 MVA base: Snom = systems,” IEEE Trans. Power Syst., vol. 27, no. 4, pp. 1818–1830, Nov.
2012.
2 p.u.; Edcnom = 2.0 p.u.; G0 = 0.06 p.u.; R = 1e-3 p.u.; X = [14] N. Trinh, M. Zeller, K. Wuerflinger, and I. Erlich, “Generic model of
5e-3 p.u.; Hc = 0.021; Hi = 7.5e − 5; Bf ilt = 0.6; ZLT C = MMC-VSC-HVDC for interaction study with AC power system,” IEEE
2e-3 + j2e-2 p.u. Kpe = 1.5; Kie = 10; Kpp = 0; Kip = 0.025; Trans. Power Syst., vol. 31, no. 1, pp. 27–34, Jan. 2016.
[15] Q. Hao, Z. Li, F. Gao, and J. Zhang, “Reduced-order small-signal models of
Kpω = 0.025; Kiω = 0.25; Kma = 7; Tma = 0.20. modular multilevel converter and MMC-based HVdc Grid,” IEEE Trans.
DFIG-based wind turbine: Pnom = 2.0 MW, H = 3.0 s, Ind. Electron., vol. 66, no. 3, pp. 2257–2268, Mar. 2019.
R = 37.5 m, ngb = 115, ρ = 1.225 kg/m3 , c1 = 0.22, c2 = [16] O. Anaya-Lara, N. Jenkins, J. B. Ekanayake, P. Cartwright, and M. Hughes,
Wind Energy Generation: Modelling and Control. Hoboken, NJ, USA:
116, c3 = 0.4, c4 = 0.0, c5 = 0, c6 = 5, c7 = 12.5, c8 = 0.08, Wiley, 2009.
c9 = 0.035, β = 0. WRIG: Rs = 0.0175 p.u, Ls = 0.2571 p.u, [17] C. E. Ugalde-Loo, J. B. Ekanayake, and N. Jenkins, “State-space modeling
Rr = 0.0190 p.u, Lr = 0.2950 p.u, Lm = 6.921 p.u, nps = 4. of wind turbine generators for power system studies,” IEEE Trans. Ind.
Appl., vol. 49, no. 1, pp. 223–232, Jan./Feb. 2013.
Pitch-angle controller: Kpb = 30, Kib = 3, Kβ = 5.0, Tβ = 10. [18] L. M. Castro and E. Acha, “A unified modeling approach of multi-terminal
B2B converter: Snom = 0.35∗Pnom , Hc = 2.652 ms, Edc = 2.0 VSC-HVDC links for dynamic simulations of large-scale power systems,”
p.u, G0r = G0g = 5e-4 p.u, R1r = R1g = 1e-4 p.u, X1r = 0.01 IEEE Trans. Power Syst., vol. 31, pp. 5051–5060, Feb. 2016.
[19] E. Acha, P. Roncero-Sanchez, A. Villa-Jaén, L. M. Castro, and B.
p.u., X1g = 0.1 p.u. RSC controller: Kpidr = 0.5, Kiidr = 8.0, Kazemtabrizi, VSC-FACTS-HVDC: Analysis, Modelling and Simulation
Kpvdr = 0.05, Kivdr = 10.0, Kpvqr = 0.05, Kivqr = 10.0. DC in Power Grids, Hoboken, NJ, USA: Wiley, May 2019.
link controller: Kpdc = 0.008, Kidc = 1.0. GSC controller: [20] J. Beerten, G. B. Diaz, S. D’Arco, and J. A. Suul, “Comparison of
small-signal dynamics in MMC and two-level VSC HVDC transmission
Kacg = 15, Tacg = 0.03. schemes,” in Proc. IEEE Int. Energy Conf., Apr. 2016, pp. 1–6.
AC power generating units: xd = 0.296 p.u, R = 5%. [21] J. Freytes et al., “Dynamic analysis of MMC-based MTDC grids: Use
– Control loops of the VSCs forming an MT-HVDC of MMC energy to improve voltage behavior,” IEEE Trans. Power Syst.,
vol. 34, no. 1, pp. 137–148, Feb. 2019.
arrangement: [22] K. Shinoda, A. Benchaib, J. Dai, and X. Guillaud, “Virtual capacitor
control: Mitigation of DC voltage fluctuations in MMC-based HVdc
systems,” IEEE Trans. Power Del., vol. 33, no. 1, pp. 455–465, Jul. 2017.
REFERENCES [23] M. Mehrasa, E. Pouresmaeil, S. Zabini, and J. P. S. Catalão, “Dynamic
model, control and stability analysis of MMC in HVDC transmission sys-
[1] Global Wind Energy Council (GWEC). Annual Market Update, tems,” IEEE Trans. Power Del., vol. 32, no. 3, pp. 1471–1482, Jun. 2017.
2017. [Online]. Available: http://www.tuulivoimayhdistys.fi/filebank/ [24] S. Zhu et al., “Reduced-order dynamic model of modular multilevel
1191-GWEC_Global_Wind_Report_April_2018.pdf converter in long time scale and its application in power system low-
[2] Offshore Wind Industry Council (OWIC). Offshore Wind Industry frequency oscillation analysis,” IEEE Trans. Power Del., vol. 34, no. 6,
Prospectus, 2018. [Online]. Available: https://cdn.ymaws.com/www. pp. 2110–2122, Feb. 2019.
renewableuk.com/resource/resmgr/publications/catapult_prospectus_ [25] O. C. Sakinci and J. Berteen, “Generalized dynamic phasor modelling of
final.pdf the MMC for small-signal stability analysis,” IEEE Trans. Power Del.,
[3] The Crown State. Offshore Wind, 2018. [Online]. Available: vol. 34, no. 3, pp. 991–1000, Feb. 2019.
www.thecrownestate.co.uk/en-gb/media-and-insights/seabed-notices/ [26] P. Kundur, Power System Stability and Control. New York, NY, USA:
offshore-wind/ McGraw-Hill, 1994.
[4] O. Anaya-Lara, J.O. Tande, K. Uhlen, and K. Merz, Offshore Wind Energy [27] T. Ackerman, Wind Power in Power Systems, 1st ed., Hoboken, NJ, USA:
Technology. Hoboken, NJ, USA: Wiley, 2018. Wiley, 2005.

Authorized licensed use limited to: UNIVERSITY TEKNOLOGY PETRONAS. Downloaded on June 20,2022 at 05:38:29 UTC from IEEE Xplore. Restrictions apply.

You might also like