Dfig XXXXXXXXX

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

www.ietdl.

org
Published in IET Renewable Power Generation
Received on 26th January 2012
Revised on 9th January 2013
Accepted on 20th January 2013
doi: 10.1049/iet-rpg.2012.0005

ISSN 1752-1416

State-space representation of DFIG-based wind


power plants
Guzmán Díaz1, Cristina González-Morán1, Ceferino Viescas2
1
Department Electrical Engineering, University of Oviedo, Campus de Viesques, 33204 Gijón, Spain
2
O&M, EdP Renewables Plaza de la Gesta, 2, 33007 Oviedo, Spain
E-mail: guzman@uniovi.es

Abstract: This study shows a methodology for state-space representation of large wind power plants, through models that include
the dynamics of the internal grid. The methodology presented fundamentally separates plant and controls at the stage of formation
of the model. It is shown then that most of the plant can be built up from elementary RL branches – with special clarification
about the reduction of the induction machine to an RL network – that are put together by means of a incidence matrix. The
problems and solutions of reference frame dynamics – complex because of the existence of two control frames for the
doubly-fed induction generator and different frames in the grid, one for each generator – are also discussed. The control is
separately formulated into a gain matrix, and control and plant are put together by simple matrix algebra. As a result,
detailed, large systems can be easily formed. The study shows an example in which ten generators are considered and shows
that there may appear particular dynamic features when multi-machine systems (rather than single, aggregated machine
systems) are represented.

Nomenclature Subscripts

Acronyms d, q direct and quadrature components


b base quantity
g1, …, gnN grid nodes 1 to nN
DFIG doubly fed induction generator B1, …, BnB grid branches (lines) 1 to nB
RF reference frame s, r, m stator, rotor and mutual
GSC grid-side converter J generic element j
RSC rotor-side converter
SSR state-space representation
SMIB single-machine infinite bus 1 Introduction

Variables State-space representations (SSRs) provide a compact


mathematical methodology for analysing systems consisting
of multiple inputs and multiple outputs (MIMO) related by
L, C, R inductance, capacitance and resistance a set of differential and algebraic equations (DAEs) that
t time constant t = (L/R) provide information about the system dynamics. In this
ω absolute rotational speed respect, power systems have been historically a particularly
δ angular position relative to a rotating RF representative instance of that type of systems (MIMO
Θ angular position relative to a stationary RF DAE-represented systems), yielding an abundant literature
Ψ flux linkage vector covering topics about control and optimisation based on
Σ leakage coefficient SSRs.
Te, Tm electromagnetic and mechanical torque Doubly fed induction generators (DFIGs) account for the
v, i voltage and current major part of the wind-based energy generation mix, and
x, u, y state, control and observation vectors their influence on some power systems is progressively
A, B, C, K state, input, observation,and gain matrices more pronounced, as profusely shown in the literature.
nG, nN, nB number of GUs, nodes and grid branches These issues, along with the experience gained over several
λ, φ, ψ eigenvalue, and right and left eigenvectors decades in analySing power systems from SSRs, would
Remark Boldfaced variables stand for vectors and serve as a justification to the formulation of SSR and
matrices subsequent analysis of DFIG systems. However, modelling

254 IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rpg.2012.0005
www.ietdl.org
DFIGs is not an easy task, on account of the several reference above commented papers [8–11]. Their models are
frames (RFs), control systems and linked dynamics between single-machine infinite-bus systems however, like most of
rotor and stator [1]; and the literature shows that the the published papers of DFIG SSRs . Indeed, the major
publications about SSRs of DFIGs are relatively scarce, trend in the literature has been to analyse and modify the
especially when several DFIGs are to be represented in a eigenstructure of one only machine – sometimes this
common power system. machine being the result of an accepted aggregation of all
An early, detailed SSR of a DFIG appeared in Ioannides’s the machines in the wind power plant – working against an
papers [2, 3]. In her papers Ioannides performed an infinite bus.
eigenvalue analysis of one DFIG considering only the In this respect, Tabesh and Iravani [12] defined an
electromechanical plant. The stator and rotor voltages were exhaustive model because the calculation of grid dynamics
exogenous inputs with no closed-loop control, and neither is integrated. As in their previous paper [13], the authors of
the dc-link dynamics, nor the mechanical input of the wind, [12] defined the grid by means of transfer function
or the interaction with the grid were considered. (Indeed, matrices, which are computed by applying the superposition
Ioannides’s papers were focused on the Scherbius drive for principle to provide a relation between voltages and
general applications, ‘including’ wind.) currents. (To obtain the transfer function, one generator at a
As expected, the papers following Ioannides’s showed a time must be considered – that is, current contributions of
progressive deepening into more complex models. In this other generators and the infinite bus are set to zero – and
respect, Mei and Pal [4] provided 4a description of a the corresponding voltage contribution at the other
DFIG-based wind system, where they introduced a generator terminals is calculated.) The task of defining a
two-mass model of the shaft and discussed the stability of wind park with several machines and lines seems laborious,
one DFIG connected to a strong busbar under constant but the notably outcome of [12] is that the eigenanalysis of
wind power. Interestingly, they followed a different a wind park would include the grid influence together with
approach to the modelling of the electromagnetic plant by the machine dynamics. To our knowledge, the work of
formulating the generator by means of equivalent voltage Tabesh and Iravani in [12] is the only one published so far
sources behind transient impedances – in fact, what they that describes the network dynamics and the interaction
did was to select those source voltages, e′d + je′q , as state among generators in a wind power plant (from the
variables rather than the rotor currents used in [2]. The small-signal analysis viewpoint).
formulation of the DFIG using e′d + je′q adhered in this way We propose in this paper an alternative approach to form
to the generally accepted procedure for formulating the SSR with detailed and joint dynamics of both the DFIG
synchronous machines. and the farm grid. Our proposal is based on a previous
Wu et al. [5] published that same year a paper in which work about islanded microgrids that we published in [14].
the SSR was improved. They retained the formulation of We postulated therein that the microgrid could be
the machine plant employing e′d + je′q as states, and also as represented in the state-space for small-signal analysis on
in [4] they reproduced a 5th-order model of the induction the basis of pattern repetitions of simple elements. Now,
machine (IM). They then considered the dynamics of following the approach in [14] we will show in this paper
the wind turbine and of the dc-link, and formulated in that a convenient division (decoupling) of the problem into
detail the control of grid-side converter (GSC) and plant and control – and also a division of the plant into
rotor-side converter (RSC). The control formulation added even more simpler sub-plants, down to a level in which for
a higher degree of accuracy in the representation of the instance only elementary RL branches remain to be
system dynamics, but because they made it in closed loop modelled – serves as well to efficiently model large-sized
the system also gained in complexity at the stage of wind power plants with reduced complexity burden. Still,
formation. we emphasise that the procedure shown in this paper does
In [6], Mei and Pal upgraded their previous representation not entail a simplification of the model; on the contrary, the
[4], and although they opted for not representing the dc-link modelling procedure includes important dynamic details for
(neither control nor dynamics) they introduced however a the study of the wind power plant – for instance, the IM is
novelty when they implemented power flow equations to modelled as a 5th order element, the dynamics of the grid
account for the DFIG–grid interaction. Those equations lines are considered, oscillations among system references
allowed one to gain insight into the effect that DFIGs are modelled, cascaded control loops are also modelled and
provoke on the network, but as we will later explain they so on. This level of detail is achieved by focusing on
are not valid to represent the dynamic response of the grid currents as the state variables that represent most of the
within a wind power plant. plant. In that sense, our approach is closer to [2, 3, 15] than
Recently, Yang et al. [7] published a paper in which a to any of the other works discussed above, but we extend
highly detailed DFIG representation was shown. Again the the electromagnetic plant to be the whole wind power plant
model was based on voltage sources behind transient by defining a pattern (the elementary RL branch) and the
impedances; and the control, truly detailed, included rules to interconnect elements based on that pattern to form
cascaded loops, feedforward terms and cross-coupling the whole farm plant.
compensations (and both RSC and GSC were considered). The paper is structured as follows. Section 2 describes the
The network was also represented (they performed a plant SSR (see Fig. 1). Special emphasis is given in that
controller optimisation in a system with two DFIGs Section to the simplification of the IM representation and its
connected to a nine-bus system), though there is not enough consonance with the modelling of farm grid lines. Section 3
information on how they modelled it, and their relevant describes the key aspects of the control formulation – with
work seems to be based again in the use of admittance some of them replicated from [14] – in also a structured
matrices and conventional power flow calculations to way. Section 4 exploits the advantages of the proposed
investigate the interactions among machines. representation – easiness of formation and sparsity – to
Other works on DFIG-based SSRs have been published show a brief small-signal analysis of a medium-sized wind
that incorporate some of the features highlighted in the power plant.

IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264 255
doi: 10.1049/iet-rpg.2012.0005 & The Institution of Engineering and Technology 2013
www.ietdl.org

Fig. 2 Equivalent circuit of a DFIG including the motion-induced


voltages that result from rotation of stator and rotor RFs

with stator and rotor flux linkages being

cs = Ls i s + Lm ir (5)

Fig. 1 Scheme of a DFIG connected to node 10 of a line with cr = Lm is + Lr ir (6)


non-negligible dynamics
Similar DFIGs are also injecting power into the line through nodes 1 to In order to put (3)–(4) in the form of (2), the flux linkages
9. Node 0 corresponds to the infinite busbar must be eliminated from the expressions. The most
straightforward procedure is using (5)–(6) in (3)–(4), and
thus the rotor and stator currents are retained as the state
2 Plant model
variables (as in the DFIG representation of [8]). This
2.1 Electromagnetic components of the plant operation introduces however an undesirable coupling
between stator and rotor currents that hinders the derivation
In [14] we demonstrated that a complex islanded microgrid of the equations in the form of (2).
could be modelled to a high level of dynamic detail from We will therefore follow another approach to get to (2). After
elementary RL (and RC) branches. To that purpose, we some manipulations, the expression for the stator voltage in
defined the complex-valued function an arbitrary RF dqa (see Fig. 3) can be recast as follows
   
m(t, v) = − t−1 + jv (1) di (a) du
v(a)
s = Rs i (a)
+ L ss
s
s
+ j a i (a)
dt dt s
where t = L/R is the branch constant time and ω the speed of a  
rotating RF in which the dynamic response is observed. Thus, di (a) dua (a)
+ Lm m
+j i (7)
the dynamics of the current i of an elementary RL branch dt dt m
driven by voltages vi and vo, defined as 

(a)
vms
 
dv
vi − vo = Ri + L + jv i whereas the rotor voltage in other arbitrary RF, dqb, reads
dt
  
could be rearranged as di (b) d ub − ur (b)
v(b)
r = Rr i (b)
+ Lsr
r
r
+j ir
dt dt
di   
= m(t, v)i + L−1 vi − L−1 vo (2) di (b) d ub − ur (b)
dt + Lm m
+j im (8)
dt dt
The elements that are eligible to be modelled in the form of 

(2) will be the IM windings, the filter-linking GSC and grid (b)
vmr
and the farm grid lines. Their modelling in accordance to
(2) is the subject of the next two subsections.

2.1.1 Induction machine: The machine can be described


using the Steinmetz circuit in Fig. 2, where particularly (i) the
rotor is fed by the RSC voltage, vr, and (ii) two flux sources
are purposely added to represent the motion-induced
voltage because of the rotation of RF at an arbitrary speed
ω. This implies that the machine dynamics are expressed as

d cs
vs = Rs i s + + jvcs (3)
dt
dcr  
vr = Rr i r + + j v − vr cr (4)
dt Fig. 3 DFIG RFs

256 IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rpg.2012.0005
www.ietdl.org
We defined in (7)–(8) the auxiliary voltages v(a) ms and vmr .
(b)
that does not affect accuracy and prevents that the three
In so doing, (7)–(8) can therefore be recast in the form of differential equations of (2) are linearly dependent.
(2); they represent the dynamics
 of branches
  Rs Lss (b)
and  The representation by means of the structure in (11)–(12)
Rr Lsr driven by voltages v(a) s − vms
(a)
and v(b)r − vmr , permits a prompt formulation of the system. It structures the
respectively. However, the more serious deficiency in this differential equations in (11) in an easy format, by
representation lies in the absence of an evident coupling separating the algebraic expression in (12) from the
between branches – the four terminal voltages are different differential equation system. Notwithstanding the expanded
and besides the equations are referred to different RFs as formulation, the order of the electromagnetic model of the
indicated by the bracketed superscripts. DFIG represented by (11)–(12) is 6, because the states
To achieve a proper coupling, we rearrange the equations involved are the six currents – the three vector currents that
(a) −jdab
by first noting the transformation i (b) m = im e , with are expanded into their d- and q-components, thus resulting
δab = θb − θa, which allows us to express the second in six currents.
auxiliary voltage now in the RF dqa It is important to note that in (11) the first and third rows are
defined in the dqa RF, whereas the second (the rotor) is
v(a) (b) jdab defined in the dqb frame. This double orientation – that is,
mr = vmr e
   this two dq-axes definition – will allow rotor and stator
di (a) ddab (a) d ub − ur (a) controls to be separately performed. This is of special
= Lm m
−j i +j im
dt dt m dt interest in DFIG control. For instance a widely accepted
  strategy aligns the rotor RF with ys and the stator RF with vs
di (a) dua (a) dur (a) [16, 17], which can approximated by δab = (π/2) rad. In
= Lm m
+j i −j i (9) such a case, expression (11) reads
dt dt m dt m
⎛ ⎞
di (a)
By comparison of (9) with v(a)
ms , it can be readily inferred that
s
⎜ dt ⎟ ⎛   ⎞
⎜ (b) ⎟ m tss , v a 0 0
⎜ di ⎟ ⎜   ⎟
dur ⎜ r ⎟=⎝ 0 m tsr , vb − vr vL ⎠
mr = vms − j
v(a) (a)
L i (a) (10) ⎜ ⎟  r m 
dt m m ⎜ dt ⎟
⎝ (a) ⎠ 0 0 m tm , va
di m
What is more interesting is that this procedure divides the dt
Steinmetz circuit in Fig. 2 into three well-defined RL ⎛ (a) ⎞ ⎛ −1 ⎞⎛ (a) ⎞
branches is Lss 0 −L−1 vs
 (a) (Rs Lss ,(b) Rr Ls(b)r , and Lm ) (a)
driven by voltages
⎜ ⎟ ⎜ −1
ss
−1 ⎟⎜ (b) ⎟
vs − vms , (a)
vr − vms , and vms ; with currents × ⎝ i (b)
r ⎠ + ⎝ 0 Lsr −Lsr ⎠⎝ vr ⎠
i (a) i i m . Additionally, the flux-dependent sources L−1
(b) (a)
s , r , and i (a)
m 0 0 m v(a)
ms
in Fig. 2 have been eliminated. Therefore the three required (13)
equations can be written in the form of (2) as follows
⎛ ⎞ Other strategies are based on aligning the rotor RF with vs
di (a)
s instead [18, 19]; again the compact representation in (11) is
⎜ dt ⎟ ⎛ mt , v  ⎞ advantageous, because the adaptation of the plant is
⎜ (b) ⎟ ss a 0 0
⎜ di ⎟ ⎜   ⎟ straightforward – db would be aligned with vs meaning that
⎜ r ⎟=⎝ 0 m tsr , vb − vr jvr Lm e−jdab ⎠
⎜ ⎟   δab = 0.
⎜ dt ⎟
⎝ (a) ⎠ 0 0 m tm , v a
di m 2.1.2 GSC filter and grid: When compared with the
dt complexity originated by the two moving elements of an
⎛ (a) ⎞ ⎛ −1 ⎞⎛ (a) ⎞ IM, the GSC filter model is of easier adaption to this
is Lss 0 −L−1ss vs
⎜ (b) ⎟ ⎜ −1 −1 −jdab ⎟⎜ (b) ⎟
structured model. It suffices to write (2) after substituting
× ⎝ i r ⎠ + ⎝ 0 Lsr −Lsr e ⎠⎝ vr ⎠ t = (La/Ra), ω = ω1, and L = La. By the same token, each RL
−1 branch in the grid – and every possible static load – could
im(a)
0 0 Lm v(a)
ms
(11) be modelled so simply. However, the next issue is how a
(large) number of isolated RL branches can be gathered in a
state matrix. In the IM network of the previous subsection
where
the coupling among stator and rotor was manifested in the
term jωr(Lm/Lσr), and the interconnection of the three RL
Lss L sr
tss = and tsr = branches was made by means of the KCL in (12).
Rs Rr However, the number of possible RL elements in the farm
grid makes it difficult – or at least cumbersome – to define
(Note in (11) that tm → ∞ because Rm ≃ 0, and thus m(tm, by hand the Kirchhoff laws (voltages and currents) at each
ωa) = − jωa.) possible node; the most convenient way to work it out
In order to be correctly formulated the system in (11) must being a matrix that reflect the system topology.
still be augmented by one equation, because an additional To that purpose we have resorted to the incidence matrix,
algebraic variable vms has been introduced. We then Γ, exactly as we did in [14] to connect all RL elements in
complete the model by introducing the algebraic expression the network. Thus, for each branch joining node j and node
k and carrying current i B jk the SSR can be put in the form
R−1
shunt vms = i s + i r e
(a) (a) (b) jdab
− i (a) (12)
m
di B jk    T
= m tB jk , v1 i B jk + L−1
B jk G vg1 · · · vgnN (14)
where Rshunt is a ‘dummy’ resistance of sufficiently high value dt

IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264 257
doi: 10.1049/iet-rpg.2012.0005 & The Institution of Engineering and Technology 2013
www.ietdl.org
As an example of the achieved normalisation of the DFIG
plant to be consonant with the grid representation, it can be
now observed that indeed there is not difference between
the representation of a grid consisting of three
wye-connected RL branches and that of a stalled DFIG (ωr
= 0) when stator and RFs are equal (and thus δab = 0).
In such a case, we can operate on (11) to yield
⎛ ⎞
di (a)
s
⎜ dt ⎟ ⎛ mt , v  ⎞
⎜ (a) ⎟ 0 0
⎜ di ⎟ ⎜ ss a
  ⎟
⎜ r ⎟=⎝ 0 m tsr , v a 0 ⎠
⎜ ⎟  
⎜ dt ⎟
⎝ (a) ⎠ 0 0 m tm , v a
di m
dt
⎛ (a) ⎞ ⎛ −1 ⎞ ⎛ (a) ⎞
is Lss 0 0 vs
⎜ (a) ⎟ ⎜ −1 ⎟ ⎜ (a) ⎟
× ⎝ ir ⎠ + ⎝ 0 Lsr 0 ⎠G⎝ vr ⎠
im
(a)
0 0 L−1
m v(a)
ms
(15)
Fig. 4 Grid RFs
where
⎛ ⎞ voltage with respect to a stationary RF αβ. This implies that
1 0 −1 dvsj is a state variable defined as
G = ⎝0 1 −1 ⎠
0 0 1 ddvsj duvsj du1
= − = vvsj − v1 (17)
dt dt dt
Thus, taken row by row (15) is equivalent to (11).
where ω1 = 100π rad/s in a 50 Hz system.
2.2 Dynamics of the RFss
2.3 Dc-link and mechanical components of the
In Section 2.1.1 the electromagnetic plant representation of an plant
IM was obtained in two generic rotating RFs: dqa for stator
and dqb for rotor. This double orientation left the plant The last task in representing the plant is the mechanical
model ready to be exploited in a next section, where the subsystem and the dc-link. Here, the pursued structuration
control will be performed by orienting the GSC and RSC is not achieved, for the equations involved cannot be
RFs along vs and ys to attain a proper control action. referred so intensively to minimal components later joined
However, still the problem needs further attention to RFs to by matrix algebra – such as it was the case with the RL
account for interactions among different DFIGs in a grid. elements. Fortunately, their repetition is not too high when
When all DFIGs are together connected to the same strong compared to the rest of elements in the wind power plant,
busbar the reference is given by the synchronising and therefore the system representation by means of the
frequency, and a common RF can be used in which the vs proposal in this paper is still favoured.
of each DFIG would be oriented with the busbar voltage, For the mechanical subsystem, the dynamic equation for
v∞. However, if lines with non-negligible impedance every DFIG is
separate the different DFIGs, each DFIG must be endowed 3
with its own RF rotating differently from v∞. dvr rAvw Cp
J = − Te − F vr (18)
This section discusses the RF associations using the scheme dt 2vr
in Fig. 4 that represents the RFs for a two-DFIG system.
Whereas the loads and grid dynamics are referred to a As in [1] we consider the RSC control aligned with ψs. Thus
common RF, each DFIG has its own RF. In Fig. 4 they are
dq∞ (grid), and dqvs1 and dqvs2 (DFIGs). (We align the RF 3pLm  
with the stator voltage for every DFIG. Therefore for DFIG Te = c i (19)
Lm + Lss s rq
number 1 the da axis in Fig. 3 corresponds to dqvs1 in
Fig. 4; for DFIG number 2, da in Fig. 3 is dqvs2 in Fig. 4; Finally for the dc-link, the power balance between capacitor,
and so on.) This means that the stator voltage of each DFIG RSC and GSC yields
must be rotated to synchronise with the grid reference.
Therefore in (11) a connection of a DFIG to node ith is dvdc
established if the following expression is supplemented for Cdc vdc = va i a − vr i r (20)
dt
the jth DFIG connected to node k

vsj = vgk e
jdvsj
(16)
3 Control representation
Thus far, we have represented the plant through a set of
We now introduce the change of variable dvsj = uvsj − u1 , structured equations fundamentally based on minimal,
where θ1 is the angular position of the infinite busbar elementary RL components – joined by means of matrix

258 IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rpg.2012.0005
www.ietdl.org
algebra through the incidence matrix – plus a reduced set of
unstructured (18)–(20). Equations (11)–(17) manifestly
represent the major portion of the wind power plant
modelling task. They introduce a set of state variables
(inductance currents, capacitor voltages, rotor speeds and
RF angles) and a set of algebraic variables (mainly node
voltages).
After introducing the linearisation terms for RL elements
(into (11)–(14)) and for RF rotation (into (16)) deduced in
[14], these equations are readily put in matrix form
ẋ = Ax + Bu. It turns out thereafter, as we demonstrated in
[14] for the case of an islanded microgrid, that now the
system (the DFIG-based wind power plant herein) can be
easily upgraded, augmented or modified by simple element
stacking each time the farm topology or composition is
modified. Besides, the interconnections are restricted to the
incidence matrix, what makes it possible to easily
reconfigure the topology.
The procedure we employed is the same as that in [14,
Section III.A]. We put the ‘open-loop’ control system in the
matrix form u = Ky, where u is the vector of exogenous
inputs, K a suitable gain matrix, and y a vector of selected
states. The topology of the control system will be restricted
then to K if we choose its elements so as to represent u =
Ky. Matrix K contains thus the information about how each Fig. 5 Control schemes
control action is exerted – that is, it informs about what From top to bottom: GSC d-axis (vdc), GSC q-axis (Qa), RSC d-axis (Qg), and
states (and other controls) affect each control variable, RSC q-axis (Te through irq). Only a PI is expanded (top scheme) so that the
despite of those states being fed through either gains, inner variables are shown. All the other PIs follow the same structure
feedforward or cross-coupling compensations. The inference
of the procedure is rather lengthy, however; so we resort to
compensation), and so on into the corresponding entries of
commend the reader to [14] for further information.  T
K, so that u = Ky hold, with u = v∗ad , v∗aq , . . . and
3.1 GSC and RSC control  T
y = eiad , iaq , jiad , . . . .
We have implemented a control of the GSC and RSC based Moreover, the separation control and plant formation
on the schemes in Fig. 5 [1]. The loops are relatively shows advantageous, because it permits us to concentrate
complex, with cascaded loops (for instance v∗ad is set by i∗ad , on the control topology, make prompt changes therein,
which turns out to be set by v∗dc ), feedforward terms (such and even use K as an unknown to synthesise optimal
as (ω2/1 + σs)|ψs|), and cross-coupling compensation terms controls [14].
(such as v1 L′a iaq when regulating iad). The procedure was
the same that we employed in [14, Section III.A]. 3.1.3 Loop closure: All loops are closed simultaneously.
In what follows, we propose to structure the control The control law u = Ky is substituted into ẋ = Ax + Bu,
formation in a three-step procedure that renders the and by means of matrix algebra the closed-loop system is
formulation simple. The steps are essentially the same that readily obtained. Caution must be taken however to
we employed in [14, Section III.A] to form the correctly form y, because the cascaded loops in Fig. 5 entail
decentralised control of an islanded microgrid in a that for instance eiad depends on a state, iad, and a control
structured, simple way. coming from an outer loop, i∗ad . A detailed description
about how to form y in case of cascaded loops can be
3.1.1 State augmentation: Note in the expanded version found again in [14].
of the top control scheme in Fig. 5 that the integrator of a PI
requires that ξiad be considered a state variable. Therefore, we 3.2 PLLs
augment the state vector with as much components as PIs in
the wind power plant. Consequently, the system matrix is also Finally, the PLLs required to orientate the RFs are modelled
augmented with terms of the form (dξ/dt) = e. No linearisation following the scheme in Fig. 6. The scheme is generic,
terms are required, and again the augmentation is conducted meaning that xα + jxβ stands for any stationary vector. The
by diagonally stacking. two orientations of axis in Fig. 3 are possible by employing
to PLLs: in our case one for ψα + jψβ (RSC control
3.1.2 Formation of the control topology: We aim at orientation) and a second for usα + jusβ (GSC control
putting the ‘open–loop’ control system in the matrix form u orientation).
= Ky, where u is the vector of inputs, K a suitable gain We represent the PLLs like control systems (as in previous
matrix and y a vector of selected states. The topology of subsection). This means that a state ξPLL must be introduced
the control system will be restricted then to K if we for each PLL in the wind power plant and that the loop is
choose its elements so as to represent u = Ky. That is, described into matrix K.
following the top scheme of Fig. 5 we observe that The output obtained from each PLL will be the sought
v∗ad = kPiad eiad + v1 L′a iaq + kIad jiad . We therefore introduce orientation. For instance, v in Fig. 6 is equal to vvs2 in
the values kP (a gain), v1 L′a iaq (a cross-coupling Fig. 4 for the DFIG number 2.
iad

IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264 259
doi: 10.1049/iet-rpg.2012.0005 & The Institution of Engineering and Technology 2013
www.ietdl.org
[We point out that the simulation times substantially differ.
The solution to (21) can be efficiently solved through
matrix notation in Matlab after defining a vector x(t)
comprising all the states. By contrast, Simulink-based
simulation of a single-machine infinite bus (SMIB) may
incorporate high frequency events not modeled in the
state-space model. This is at the cost in our experience of
simulation times two order of magnitude higher.].
Fig. 6 Generic PLL model On this basis, Fig. 7 shows a comparison of the transient
The input is a variable in stationary RF, x = xα + jxβ. The output is the simulation with the plots obtained from the proposed SSR,
rotational speed ω of the RF that makes xq = 0 and thus aligns the RF to x which shows that the modes that build the transient
by yielding |x| = xα. The angle θ is obtained from ω and used as a feedback response are almost equal by the two methods. Certainly in
to transform x from the stationary RF to the rotating RF
the three plots it is observed a discrepancy in the reach of
the signals. (It is most noticeable over the first 0.5 s of Tm,
4 Numerical results where the plot from the eigenanalysis-based data is higher
than from Simulink. It is also noticeable in Te, where the
4.1 Validation oscillations are larger in Simulink-based data than in
eigenanalysis-based data [In addition to the spikes in the
Small-signal analysis differs from transient simulation signal, which are related to modes of the PWM
analysis in that this latter solves the system differential high-frequency switching that we have not modelled in the
equations to yield the time evolution of the variables (using SSR.].) However, the explanation for this behaviour lies not
packages like DIgSILENT or Simulink), whereas by in the formation of the SSR but in the linearisation. Fig. 8
contrast the small-signal analysis plots the time evolution shows the plot ωr against Tm, obtained through the direct
using a prior eigenanalysis of the linearised jacobian. processing of Fig. 7. Importantly the representation is
Through the eigenanalysis, a system of order n is equivalent to a portion of the turbine mechanical
represented by its n modes (consisting of eigenvalues, λi, characteristic – the precise torque is represented as the thin,
and right and left eigenvectors, φi and ψi, i = 1,…, n), grey line of Fig. 8, which was obtained by mixing data
which are summed to plot the instantaneous value of a state from Fig. 7, and it can be accurately overlaid with the plot
variable in the form [20, §12.2] of the mechanical characteristic. At ωr = 1.025 pu upon
which the wind is increased, the linear approximation of Tm

n   versus ωr observed in Fig. 8 makes it possible for the
x(t) = fi ci x(t0 ) eli t (21) eigenanalysis-based computation to top higher in Fig. 7
i=1 than the precise mechanical torque. As the rotor speeds up,
the agreement between both curves in Fig. 8 improves, up
Transient-based and eigenanalysis-based methods differ to the operation point employed for linearisation
completely in the way in which the data (to be plotted) are (vr ≃ 1.2 pu) where the coincidence between both
obtained, and therefore the agreement between both Simulink-based and eigenanalysis-based is evident. As
methods of analysis can be used as a validation method expected, at that operation point the eigenanalysis-based

Fig. 7 Evolution of torque and rotor speed for an increase of wind speed from 10 to 11 m/s
Thin, grey lines: transient simulation (Simulink); thick, black line: from proposed SSR

260 IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rpg.2012.0005
www.ietdl.org

Fig. 8 Plot of ωr against Tm over the interval of 1.5 s represented in Fig. 7


Thin, gray line: non-linearised Simulink; thick, black line: linearised SSR

approximation is tangent to the Simulink-based observed in the electrical torque, Te, which is an algebraic
approximation. variable depending on the machine flux. Also more damped
On the whole, where the tangent does not separate in but remarkably oscillatory as well are the modes
excess from the curve the approximation is good, and it corresponding to the GSC current (no. 19). The rest of
loses accuracy otherwise for large perturbations. It is a modes are notably damped, with a slow damped mode
general, undesirable feature of small-signal analysis – corresponding to the rotor speed (no. 15) that explains the
intended to summarise valuable information of the system evolution of ωr in Fig. 7.
dynamics into its Jacobian, but at the cost of linearising the
system around an operating point. Eventually, this turns out
to be the cause for some distortions to be expected when 4.3 Eigenanalysis of several machines
comparing results from large perturbations such as the
example presented here. Mindful of this limitations, the The previous example was an instance of a single-machine
comparison of transient-based and eigenanalysis-based data infinite-bus system analysis, which has been already
must be prudently interpreted when the signal separates in covered in detail in the related literature [8–11, 15]. Now
excess from the operating point. On that account, the we show here by means of a simple example that the
analysis of the results in Fig. 7 fundamentally serves as a consideration of several DFIGs (joined by lines with
validation of the structured SSR . non-negligible impedances) rather than the use of one
single machine does affect the final results, and that the
accuracy may be compromised in some scenarios when a
4.2 Eigenanalysis of one only machine simple aggregation (summation) of several machines into
one only DFIG is performed.
Fig. 7 shows the transient response of one only machine – that In Fig. 10 we show the results for two configurations. A
connected to node 1 in Fig. 1, close to the infinite bus – when first configuration consists of one only machine in node 10
the rest of the generators are not considered. The spectrum for (see Fig. 1), indicated with asterisks. The second
that case is represented in Fig. 9. Additionally, Table 1 shows configuration considers 10 equal DFIGs connected to nodes
the variables with the largest participation in each of the 1 through 10, with the eigenvalues shown as crosses.
eigenvalues in Fig. 9. (Complex eigenvalues are represented Initially, for a wind vw = 10 m/s in all DFIGs both
by only one of the pair of conjugate eigenvalues, multi-machine and single-machine configurations are stable.
accounting for the symmetry.) Each participation factor is However, Fig. 10 represents the case in which the wind
generally an indicative of the relative participation of the ith acting on the DFIG in node ten drops to 8 m/s, and
mode in the kth state variable and is obtained from the additionally in the multi-machine configuration the wind of
eigenstructure of the system as the adjacent DFIG number nine falls to 7.5 m/s (all the
others DFIGs maintained at vw = 10 m/s). The oscillations
pki = fki cik (22) of Te are now less damped (see eigenvalues indicated with
asterisks pulsating at 314 rad/s), and an analysis of
where φki is the kth element of the right eigenvector and ψik is participations shows now cross participations among the
the kth element of the left eigenvector ψi, both corresponding torques of the machines in the multi-machine configuration.
to the ith eigenvalue, λi [21]. This is an issue observed also in the stable multi-machine
Inspection of the data shows undamped representative configuration – that is, the eigenvalue no. 16 in Fig. 9
egigenvalues (no. 16 at − 0.51 ± j314 rad/s) that are moves to the right when additional DFIGs are connected
highly participated by the magnetising current im. This in nearby; the rest of the damped eigenvalues moving less
turn is the reason for the oscillation at the grid frequency noticeably.

Fig. 9 Eigenvalues for a single-machine system obtained after connecting a unique DFIG to node 1

IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264 261
doi: 10.1049/iet-rpg.2012.0005 & The Institution of Engineering and Technology 2013
www.ietdl.org
Table 1 Summary of the eigenvalues with the largest wind speed (see zoomed area in Fig. 10), the multi-machine
participation in related state variables corresponding to Fig. 9 model yields a pair of complex conjugate eigenvalues in the
Eig. State Particip. right-half plane. The modes are highly participated by the
variables vdc and ξdc of the DFIG number 9. The modes
1 iBq 0.98 are so dominant that they lead to a characteristic loss of
2 iBd 1.00 stability that would not be otherwise represented by the
3 ird, isd 0.67, 0.50 single-machine model.
4 irq, isq 0.56, 0.34 Fig. 11 further clarifies the stability nature of both
5 irq, isq 0.46, 0.65
6 ird, isd 0.34, 0.47 representations, by representing the time evolution of the
7 ξQ(RSC) 1.05 state variable iqr, proportional to the electromagnetic torque
8 dvs 1.05 as reflected in (19).
9 ξiad, ξiaq 0.52, 0.52 Both figures show the response of the state variable to a
Eig. State Particip.
10 ξPLL 1.05
perturbation in the wind speed that varies its value from 9
11,12 ξird, ξirq 0.55, 0.45 to 10 m/s. Fig. 11a shows the loss of stability derived from
13 vdc, ξdc 1.93, 0.93 the existence of the real eigenvalue in Fig. 10 when a
14 vdc, ξdc 0.93, 1.93 distributed representation is chosen. By contrast, though the
15 ωr 0.99 underlying high-frequency mode is approximately the same,
16 imd, imq 0.49, 0.50
17,18 iL0, iL1 0.5, 0.5 the response of the aggregated model exhibits a damped
19 iad, iaq 0.52, 0.52 oscillation that leads the system to a stable point.

4.4 SMIB and distributed DFIGs

We finally discuss some particular results observed when we


compared the performance of a wind power plant with that of
an equivalent SMIB, widely employed in the related
literature.
Fig. 12 shows the value of the real part of the eigenvalue
that is closest to the right-half plane, max{ℜ(λi)}, for
different wind speeds. We employed an SMIB that
Fig. 10 Eigenvalues of single- (asterisks) and multi-machine consisted in four equal DFIGs connected to node 1 in
(crosses) systems Fig. 1. We then obtained the upper plot of Fig. 12 by
The data have been obtained from an scenario in which the wind speeds were
conducting an eigenanalysis of the SMIB system when the
8 m/s for DFIG in node 10 (in both multi- and single-machine wind speed ranged from 7.5 to 18 m/s; with all four DFIGs
configurations), 7.5 m/s for DFIG in node 9, and 10 m/s for the other eight subject to the same wind speed because we wanted and
DFIGs equivalent aggregated machine. For the bottom plot, we
considered a wind power plant with four machines now
distributed in nodes 2, 5, 9 and 11. With this second model
The most remarkable result nevertheless is the loss of we obtained the eigenstructure in 100 cases, in which the
stability in the multi-machine configuration. Whereas the four machines were subject to different wind speeds. (The
single-machine configuration remains stable with decreased distance between node 2 and 11 in the actual farm is of

Fig. 11 Time evolution of irq for multimachine system


a Aggregated model
b As obtained from the eigenvalue plot in Fig. 10

262 IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rpg.2012.0005
www.ietdl.org
elementary RL branches joined by means of convenient
graph theory techniques.
Through a simple example we have shown that the
multi-machine configuration is capable of representing local
deviations in a wind power plant that would be otherwise
difficult to describe by means of one only aggregated
machine that were the sum of all the DFIG capabilities. The
results from the example indicate then that the aggregation
should be carefully conducted – we suggest that through
specific techniques (see [22] and references therein) – by
processing the SSRs of the wind power plant. Also we
point out that the system analysed in this paper is really
simple (Fig. 1), but we also emphasise that larger systems
(wind power plants) can be more in depth analysed thanks
to the easiness of system augmentation following the
Fig. 12 Values of the real part of the eigenvalue closest to the
approach proposed in this paper (see for instance the
right-half plane as a function of wind speed
69-node grid in [14]).
Top: one only generator; bottom: four generators

6 Acknowledgments
12.4 km, thus justifying different wind speeds.) We employed
a Weibull distribution of parameters 13 and 2.1 to generate a This work was supported in part by the Spanish Ministry of
vector of stochastic wind speeds. To plot the results (bottom Science under Grant ENE2010-14899 and in part by EdP
of Fig. 12) we employed the average wind speed. Renewables under Grant FUO-EM-315-10.
We observe two differences between the SMIB and the
multi-machine models that raise more concerns about the
direct use of simple aggregated machines (SMIBs) in 7 References
detailed eigenanalyses of wind power plants. First, the max
{ℜ(λi)} is in general farther from the right-half plane at 1 Pena, R., Clare, J.C., Asher, G.M.: ‘Doubly fed induction generator
high wind speeds for the SMIB; with a clear trend – the using back-to-back PWM converters and its application to
variable-speed wind-energy generation’, IEE Proc. Electr. Power
higher the wind speed, the more stable the system. This Appl., 1996, 143, (3), pp. 231–241
seems to be a consequence of the line impedance. (Indeed, 2 Ioannides, M.G.: ‘Doubly fed induction machine state variables model
if we were to find a SMIB less stable, we could increase and dynamic response’, IEEE Trans. Energy Convers., 1991, 6, (1),
the impedance between the aggregated machine and the pp. 55–61
infinite bus – which in our analyses has shown to move 3 Ioannides, M.G.: ‘State space formulation and transient stability of the
double output asynchronous generator’, IEEE Trans. Energy Convers.,
the max{ℜ(λi)} to the right.) Second, there is a clear 1993, 8, (4), pp. 732–738
dispersion in the value of max{ℜ(λi)} when the average 4 Mei, F., Pal, B.: ‘Modal analysis of grid-connected doubly fed induction
wind speed is used to compare both models, which hides generators’, IEEE Trans. Energy Convers., 2007, 22, (3), pp. 728–736
the clear trend observed in the SMIB model. Thus, points 1 5 Wu, F., Zhang, X.P., Godfrey, K., Ju, P.: ‘Small signal stability analysis
and optimal control of a wind turbine with doubly fed induction
and 2 correspond to a wind speed of 12 and 13 m/s and generator’, IET Gener. Transm. Distrib., 2007, 1, (5), pp. 751–760
are located at −15.6 and −17.5 mrad/s, whereas for the 6 Pal, B.C., Mei, F.: ‘Modelling adequacy of the doubly fed induction
similar average wind speed the multi-machine model generator for small-signal stability studies in power systems’, IET
gives are various: ( −15.6, −12.4, −7.3, −7.4, −1.6) Renew. Power Gener., 2008, 2, (3), pp. 181–190
mrad/s [points 3 through 7, with wind speeds 7 Yang, L., Yang, G.Y., Xu, Z., Dong, Z.Y., Wong, K.P., Ma, X.:
‘Optimal controller design of a doubly-fed induction generator wind
(12.98,12.75,13.24,13.20,12.48) m/s]. In all cases the turbine system for small signal stability enhancement’, IET Gener.
participation analysis showed that the most participated Transm. Distrib., 2010, 4, (5), pp. 579–597
variable was always the rotational speed – as eigenvalue 18 8 Lingling, F., Kavasseri, R., Zhixin Lee, M., Chanxia, Z.: ‘Modeling of
in Fig. 9 – but the responsible DFIG was different: for DFIG-based wind farms for SSR analysis’, IEEE Trans. Power Deliv.,
2010, 25, (4), pp. 2073–2082
points 3 through 7 the variables most participated by the 9 Ostadi, A., Yazdani, A., Varma, R.K.: ‘Modeling and stability analysis
corresponding max{ℜ(λi)} were respectively ωr2, ωr11, ωr9, of a DFIG-based wind-power generator interfaced with a
ωr5, and ωr5, with the subscript showing the number of the series-compensated line’, IEEE Trans. Power Deliv., 2009, 24, (3),
node at which the DFIG most participated was connected. pp. 1504–1514
10 Mishra, Y., Mishra, S., Li, F., Dong, Z.Y., Bansal, R.C.: ‘Small-signal
stability analysis of a DFIG-based wind power system under different
modes of operation’, IEEE Trans. Energy Convers., 2009, 24, (4),
pp. 972–982
5 Conclusions 11 Rahimi, M., Parniani, M.: ‘Dynamic behavior analysis of doubly-fed
induction generator wind turbines – the influence of rotor and speed
We have presented in this paper a methodology for devising controller parameters’, Int. J. Electr. Power Energy Syst., 2010, 32,
detailed SSRs of wind power plants in a structured way. (5), pp. 464–477
The procedure is based on separating control and plant at 12 Tabesh, A., Iravani, R.: ‘Small-signal model and dynamic analysis of
variable speed induction machine wind farms’, IET Renew. Power
the stage of formation of the SSR; that are put together by Gener., 2008, 2, (4), pp. 215–227
matrix algebra when the system open-loop matrix and the 13 Tabesh, A., Iravani, R.: ‘Small-signal dynamic model and analysis of a
gain matrix are finished. In so doing, we have found real fixed-speed wind farm – a frequency response approach’, IEEE Trans.
procedural advantages because of the simplified Power Deliv., 2006, 21, (2), pp. 778–787
14 Diaz, G., Gonzalez-Moran, C., Gomez-Aleixandre, J., Diez, A.:
representation of the electromagnetic plant – a ‘Complex-valued state matrices for simple representation of large
simplification that has been accomplished thanks to the autonomous microgrids supplied by PQ and Vf generation’, IEEE
arrangement of the DFIG equations and of the farm grid as Trans. Power Syst., 2009, 24, (4), pp. 1720–1730

IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264 263
doi: 10.1049/iet-rpg.2012.0005 & The Institution of Engineering and Technology 2013
www.ietdl.org
15 Yang, L., Xu, Z., Østergaard, J., et al.: ‘Oscillatory stability and 8.1 DFIG electromagnetic plant
eigenvalue sensitivity analysis of a DFIG wind turbine system’, IEEE
Trans. Energy Convers., 2011, 26, (1), pp. 328–339 Rs = 0.0075, σs = 0.0222, Rr = 0.018, σr = 0.0222, Lm =
16 Hopfensperger, B., Atkinson, D.J., Lakin, R.A.: ‘Stator-flux-oriented
control of a doubly-fed induction machine with and without 4.2520, Ra = 3.1 × 10−4, La = 1.1313, p = 2.
position encoder’, IEE Proc. Electr. Power Appl., 2000, 147, (4),
pp. 241–250 8.2 Aerodynamics
17 Ledesma, P., Usaola, J.: ‘Doubly fed induction generator model for
transient stability analysis’, IEEE Trans. Energy Convers., 2005, 20,  
116
(2), pp. 388–397 CP = 0.5176 − 0.4b − 5 e−(21/li ) + 0.0068l;
18 Peresada, S., Tilli, A., Tonielli, A.: ‘Power control of a doubly fed li
induction machine via output feedback’, Control Eng. Pract., 2004,
12, (1), pp. 41–57 r = 1.225 kg/m3
19 Chondrogiannis, S., Barnes, M.: ‘Stability of doubly-fed induction
generator under stator voltage orientated vector control’, IET Renew. kP = 0.4847 W s3 , kb = 10 s, and lmax = 8.1
Power Gener., 2008, 2, (3), pp. 170–180
20 Kundur, P., Balu, N.J., Lauby, M.G.: ‘Power system stability and
control. The EPRI power system engineering series’ (McGraw-Hill,
New York, 1994) 8.3 Lines
21 Perez-Arriaga, I.J., Verghese, G.C., Schweppe, F.C.: ‘Selective modal
analysis with applications to electric power systems. Part I: Heuristic See Table 2.
introduction’, IEEE Trans. Power Appar. Syst., 1982, PAS-101, (9),
pp. 3117–3125 Table 2 Line parameters (in pu, base 20 kV)
22 Marinescu, B., Mallem, B., Rouco, L.: ‘Large-scale power system
dynamic equivalents based on standard and border synchrony’, IEEE Line RB LB CB
Trans. Power Syst., 2010, 25, (4), pp. 1873–1882
0–1 0.0012 × 10−3 0.0031 × 10−3 0.0010
1–2 0.6667 × 10−3 0.2369 × 10−3 0.4126
2–3 0.1132 × 10−3 0.0402 × 10−3 0.0701
3–4 0.1703 × 10−3 0.0514 × 10−3 0.0781
8 Appendix 1. Model parameters 4–5 0.1183 × 10−3 0.0242 × 10−3 0.0291
5–6 0.0831 × 10−3 0.0251 × 10−3 0.0381
6–7 0.0494 × 10−3 0.0149 × 10−3 0.0226
All values are given in pu, with bases 1.27 MVA and 690 V 7–8 0.1508 × 10−3 0.0309 × 10−3 0.0371
8–9 0.0447 × 10−3 0.0135 × 10−3 0.0205
(LV) and 20 kV (MV). 9–10 0.1705 × 10−3 0.0349 × 10−3 0.0420

264 IET Renew. Power Gener., 2013, Vol. 7, Iss. 3, pp. 254–264
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rpg.2012.0005

You might also like