Electric Power Systems Research 105 (2013) 3950

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

Transient analysis of low-voltage ride-through in three-phase

grid-connected converter with LCL lter using the nonlinear modal
series method
Wei Wang , Xikui Ma
State Key Laboratory of Electrical Insulation and Power Equipment, School of Electrical Engineering, Xian Jiaotong University, Xian 710049, China

a r t i c l e i n f o a b s t r a c t

Article history: From a theoretical perspective, this paper develops a systematic approach to analyze the transient charac-
Received 31 January 2013 teristics of the low-voltage ride-through (LVRT) in three-phase grid-connected converter with LCL lter
Received in revised form 5 July 2013 by using the nonlinear modal series method. Firstly, the mathematical model and the corresponding
Accepted 15 July 2013
second order modal series model of the system are described. And then, the simulation results and the
Available online 17 August 2013
theoretical analysis manifest that there are abundant modal interactions, particularly the stronger second
order modal interaction, in the LVRT transient process. The revised expressions considering the nonlin-
Keywords:
ear terms for the state variables are subsequently obtained. By selecting the appropriate parameters to
Transient analysis
Low-voltage ride-through (LVRT)
weaken second order modal interaction, we can effectively reduce the amplitude and the duration of the
Three-phase grid-connected converter oscillation to satisfy the requirements of the system tolerance during LVRT process. Additionally, the dom-
Nonlinear modal series method inant oscillation modes of each state variable are also studied. Finally, the second order quasi-resonance
Modal interaction. boundaries are dened to facilitate the selection of practical parameters for maintaining normal transient
behavior.
2013 Elsevier B.V. All rights reserved.

1. Introduction
The three-phase power electronic converters with LCL lter
are now used in many renewable energy generation systems as a
grid-connected interface (e.g., wind energy, solar etc.) [1,2]. Power
regeneration, adjustable power factor and signicantly less line
current harmonic distortion are the most important advantages of
this converter with respect to other types of converters [3]. As the
scale of renewable energy generations becomes larger and larger
nowadays [4,5], the huge penetration of these new power to the
grid has led the grid codes to require the low-voltage ride-through
(LVRT) capability of the grid-connected converter in these gener-
ation systems during the grid disturbances to avoid the shutdown
phenomenon of large renewable energy generations [6,7]. As an
example, a diagram of the LVRT requirements, in which wind tur-
bines should be still connected for voltage sags, is shown in Fig. 1
[8,9]. Practically speaking, these renewable energy generation sys-
tems has been faced with serious problems during the LVRT process
and the grid code requirements about LVRT present a signicant
challenge to the well-established renewable energy technologies
[10,11]. In the power system where the renewable energy gener-
ation is of a major portion, the system variables such as the grid
Corresponding author. Tel.: +86 13572243436.
E-mail addresses: weiwang.eec@gmail.com, 240216@163.com (W. Wang).
voltage and the grid current are much uctuated in comparison
with the normal operation state during voltage dip [12], which has
an important effect on the transient stability and applications in
many areas of these power generations.
For the issues of LVRT, the previous work is mainly focused
on a suitable ride-through control method or an additional hard-
ware circuit to enhance the LVRT capability of the grid-connected
converter in the power generation system, which will allow the
operation of the renewable energy generation system to meet
the requirement of the set of grid codes [1316]. Unfortunately,
since power system faults are often characterized by a momentary
decrease in the RMS voltage magnitude [17], these works ignore
the various transient phenomenon existing in the LVRT process and
does not further analyze the transient behavior of the LVRT which
might be of a powerful impact on the grid and or even destroy the
system circuit. The investigation into transient behavior allows an
essential understanding of the system dynamical operations and
may give some useful information for practical design and control
applications [18,19].
Recently, some works detailing the transient state of volt-
age sags on the grid-connected converter in the renewable
energy generation system have appeared, and the transient phe-
nomenon has been used to assess the oscillations at external
grid voltage sag [20]. For example, Muyeen et al. have carried
out extensive time-domain simulation analyses on the transient
behavior of wind turbine system considering both symmetrical and

0378-7796/$ see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.epsr.2013.07.007
40 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950
Fig. 1. National grid codes.
unsymmetrical faults occurring at several locations to analyze the
LVRT characteristics of the system [21]. And the simulation waves
of PV converter in LVRT process are also investigated in relative
literatures [22]. Although the traditional time-domain simulation
method can get some nonlinear transient information by observing
the time-domain simulation phenomena, it is difcult to identify
the nature of system oscillations, and the relationship between
the interactive modes and the state variables, as well as system
parameters and disturbances. Essentially, the existing works have
not dealt with the effects of nonlinear modal interactions on the
dynamic behavior of the grid-connected converter in LVRT process,
and the consequent relationship between the system behavior and
the control strategies. Moreover, traditional modeling methods use
a linear model and the validity of linear analysis method such as
RouthHurwitz stability criterion is restricted to a small neighbor-
hood of the operating point which is not suitable to the case of large
disturbance [23], so that the mechanism of LVRT transient behav-
ior can not be accurately revealed. Therefore, an analysis of LVRT
transient behavior for the grid-connected converter with LCL lter
in renewable energy generation system is absent from a theoretical
perspective.
In this paper, the nonlinear modal series method, by which
the linear system theory concepts can be extended to facilitate
the understanding and analysis of nonlinear system [24] through
considering the nonlinear high-order polynomial in high-order
modes space, is used to investigate the LVRT transient dynamic
characteristics of the grid-connected converter with LCL lter
during voltage sag so as to transfer the transient characteristics
research from the qualitative analysis to the quantitative analy-
sis. The intrinsic mechanism of the way by which the nonlinear
modal interaction affects the LVRT transient behavior is revealed,
and more appropriate functions for representing system behav-
ior can be obtained. Based on the relationship between the modal
interaction and the parameters, the oscillation amplitude and dura-
tion of the state variable during LVRT transient process can be
reduced to satisfy the requirements of the system tolerance by
adjusting certain of the parameters. Furthermore, the second order
quasi-resonance boundaries with regard to some key parameters
are analyzed, in order to facilitate the optimal design of the grid-
connected converter with LCL lter.
This paper is organized as follows. In Section 2, the mathe-
matical model of a interface system constituted by grid-connected
converter with LCL lter and its control scheme are developed. In
Section 3, the second order modal approximate solution of the sys-
tem is established by using the nonlinear modal series method. In
Section 4, more nonlinear modes are found using the second order
modal method, and the simulation is also carried out to illustrate
the efciency of the nonlinear modal series method in the analy-
sis of the LVRT transient behavior. Then, a quantitative measure is
used to nd out the dominant nonlinear interactive modes for the
state variable and the appropriate formulations of the main state
variables are presented by selecting the stronger modal interaction.
The relationship between the nonlinear modal interactions and the
system parameters are newly performed to assess the parameters
impacts on the LVRT transient behavior. Subsequently, the domi-
nant oscillation modal of the system state variables are also studied.
In order to avoid the occurrence of resonance, the second order
quasi-resonance boundaries are given in this subsection. Finally,
Section 5 concludes the paper.
2. System modeling
2.1. Description and mathematical model
The grid-connected converter is commonly based on a voltage
source inverter (VSI) connected to the supply network [25]. These
converters are always used to maintain a constant DC-link voltage
and to control the power factor of the generation system. The DC-
link is created by a capacitor which isolates the operation of the
grid-connected converter from renewable energy sources, so that
the control and the operation of the converter allow being indi-
vidual. In practice, the renewable energy generation system only
needs to take countermeasures for the grid-connected converter
and the DC-link in the case of LVRT process which will not affect
the normal operations of the renewable energy source. In order to
deeply investigate the transient behavior, these generation systems
should be simplied. Here, the equivalent circuit of the system is
shown in Fig. 2. In this paper, a grid-connected interface system
constituted by a three-phase converter with LCL lter as a research
model will be discussed.
In most of these studies, the power source of the system is
considered as a DC voltage source [26]. However, this equivalent
treatment is not adaptable to the control principles of the grid-
connected converter in the renewable energy generation system
such as PMSG-type wind turbine. In addition, the input current of
the converter is not predictable. In order to establish a reasonably
alternative model, the basic principles of a simplied treatment
should ensure that both the voltage and the current characteristics
of the simplied model are as same as the prototype, and the con-
trol method of the grid-connected converter remains unchanged.
Based on these preconditions, a controlled current source is used
to replace the renewable energy source and its related power
Fig. 2. Equivalent circuit of grid-connected renewable energy generation system.
 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950 41

Fig. 3. The grid-connected interface in the equivalent circuit.

electronic devices. The current of the controlled current source is control scheme requires the measurements of the DC-link voltage,
determined by the system power and the voltage of the DC-link. the grid voltage and the grid current. The converter is controlled in
And the control method of the DC-link voltage is the same as the the grid voltage oriented reference frame, with its d-axis oriented
original renewable energy generation system. As a result, the grid- along the grid voltage vector position [27].
connected interface in the equivalent circuit is shown in Fig. 3, and Since the treatment of controller input is carried out by per unit,
here iccs = Pt /uDC . the reference value of the per unit and the transformation rela-
Here, let the direction of the current owing into the grid be tionships between the actual value and the per unit are given by
positive. Under a dq reference frame rotating at grid speed, the dif-
ferential equations for the grid-connected converter with LCL lter
2SB
can be written as IB dq = (7)
3UB
di1d R1 1 1
= i1d + i1q + u u (1)
dt L1 L1 d L1 0d Id/q Id/q
Id/q = = (8)
IB dq 2S
B
di1q R1 1 1 3UB
= i1d i1q + uq u0q (2)
dt L1 L1 L1
where SB , UB , and IB dq are the base value of the system power, the
grid voltage and the grid current, respectively. Id/q and I*d/q are the
di2d R2 1 1
= i2d + i2q + u u (3) nominal value and per unit of d-axis and q-axis current under dq
dt L2 L2 0d L2 sd
reference frame, respectively.
Usually, the feed-forward compensation does not affect the
di2q R2 1 1
= i2d i2q + u0q usq (4) accuracy of the system. On account of this fact, we use the PI
dt L2 L2 L2
controller only and ignore the feed-forward compensation in the
inner current control loop for converter control. Accordingly, the
du0d R1 RC 1 R2 RC 1 1 1
= ( + )i1d + ( )i2d RC ( + )u differential equations regarding the control of the converter are
dt L1 C L2 C L1 L2 0d expressed as follows:
RC RC
+ u0q + u + u (5) dud 1 di2dref 3UB di2d 3UB
L1 d L2 sd = (K2p K2p + K2i (i2dref i2d ))
dt 0.67uDC dt 2SB dt 2SB
(9)
du0q R1 RC 1 R2 RC 1
= ( + )i1q + ( )i2q u0d
dt L1 C L2 C

1 1 RC RC duq 1 3UB di2q 3UB
RC ( + )u0q + uq + usq (6) = (K2p + K2i (i2qref i2q )) (10)
L1 L2 L1 L2 dt 0.67uDC 2SB dt 2SB
where iL1 = i1d + ji1q , iL2 = i2d + ji2q are the L1 and L2 current vectors,
di2dref duDC
respectively; uo = uod + juoq is the voltage vectors of capacitor C; R1 , = K1p + K1i (uDCref uDC ) (11)
R2 and RC are the resistance of L1 , L2 and C, respectively; is the dt dt
grid voltage angular frequency. where u = ud + juq is the converter output voltage vector;
The generic control scheme of the grid-connected converter in i2 = i2d + ji2q and i2ref = i2dref + ji2qref are the grid current vector and
renewable energy generation system is illustrated in Fig. 4. The its reference, respectively; uDC is the DC-link voltage; K1p and K1i

Fig. 4. Control scheme of the grid-connected converter.

42 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950

are the proportional gain and the integral gain of the DC-link con- variables yields:
troller, respectively; K2p and K2i are the proportional gain and the
integral gain of the grid current controller, respectively.
The equation describing the energy balance of the DC-link
capacitor can be written as
J11 J12 0 0 J15 0 J17 0 0 0
duDC
= Po Pt = 1.5(ud i1d + uq i1q ) Pt
CDC uDC (12) J21 J22 0 0 0 J26 0 J28 0 0
dt

0 0 J33 J34 J35 0 0 0 0 0
where Po and Pt are the power supplied to the grid and generated
by the renewable energy source, respectively. 0 0 J43 J44 0 J46 0 0 0 0

Noting Eq. (12), then, Eqs. (911) can be rewritten as
J51 0 J53 0 J55 J56 J57 0 0 0
X =

X

dud 2.25K1p K2p i1d ud 2.25K1p K2p i1q uq
= 0 J62 0 J64 J65 J66 0 J68 0 0
dt CDC u2DC CDC u2DC

J71 J72 J73 J74 J75 0 J77 J78 J79 J710
3UB R2 K2p 3UB L2 K2i i2d 3UB K2p i2q
+ 0 0 J83 J84 0 J86 0 0 0 J810
u 0.67 2SB uDC
0.67 2SB L2 DC
J91 J92 0 0 0 0 J97 J98 0 J910
3UB K2p u0d K i2dref K1p K2p Pt 1
+ 2i + J101 J102 0 0 0 0 J107 J108 0 J1010
0.67 2SB L2 uDC 0.67 uDC 0.67CDC u2

DC

0
K1i K2p uDCref 1 3UB K2p usd 1 K1i K2p
+ + (13) 0
0.67 uDC 0.67 2SB L2 uDC 0.67
0

0

2
3UB K2p 2 3U L K
3UB R2 K2p 0
1
duq i2d B 2 2i i2q
= + + 0 + (17)
dt 0.67 2SB DCu 0.67 2SB L2 uDC 2
X T H7 X
2 2 u
3UB K2p u0q 3UB K2p sq 1 K2i i2qref 1 T
+ + (14) X H8 X
0.67 2SB L2 uDC 0.67 2SB L2 uDC 0.67 uDC
T
X H9 X
di2dref 1.5K1p i1d ud 1.5K1p i1q uq K1p Pt 1 X T H10 X
= + K1i uDC
dt CDC uDC CDC uDC CDC uDC
+ K1i uDCref (15)

where X belongs to the convergence domain of the Taylor series

From Eq. (1) to Eq. (15), we can obtain a set of differential equa-
tions to present the model of the grid-connected converter with
LCL lter, rewritten in a compact form as
X = f (X) (16)
where X = [i1d , i1q , i2d , i2q , uod , uoq , ud , uq , i2dref , uDC ]T is the vectors
containing the system state variables, and F : RN RN is a smooth
vector eld.
v RN , the coefcient matrix of X is the Jacobian matrix which rep-
resents the linear part of the vector eld and equals [fi /X]X=X0 , J
are the nonzero elements of the Jacobian matrix, Hi is the Hessian
Matrix for the state variable xi , where i = 7, 8, 9, 10.
In order to get the standard form of the nonlinear modal series
processing method, denote the (complex) Jordan form of A by  and
the matrices of the right and left eigenvectors by U and V, respec-
tively. Then, using the transformation X = UY yields the following
equivalent system for (17)

2.2. Nonlinear modal series

As a powerful tool, nonlinear modal series theory can yield a

1
good deal of theoretical perspective into the dynamic analysis of
nonlinear systems. Particularly, the method of solution has the 2

great conceptual advantage of presenting a nonlinear system as
3
a rather straightforward generalization of the linear case, although
4
it may be much more involved [24]. The basic idea of modal series
approach is that a Taylor-series expansion is made for the nonlin-
5
ear system dynamics equations around a stable operating point. Y =

Y

The order of the series expansion is closely related to the nature 6

of the nonlinearities in the original dynamical system. As a result,
this method provides a closed-form approximate solution to the 7

differential equation, as well as the resonance condition. 8

Suppose f is a continuous and expansible function, and the order
of the system (16) is N. Expanding (16) as a Taylor-series around 9
a stable equilibrium point X0 and using again X as the new state 10
 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950 43

Table 1
0
System parameters used in simulations.
0

0 Parameters Descriptions Values

0 R1 , L1 Inverter-side resistance and inductance of 0.004 , 1.6 mH
the LCL lter
0
1 R2 , L2 Grid-side resistance and inductance of the 0.004 , 1.6 mH
+ V 0 + (18) LCL lter
2
Y T U T H7 UY RC , C Resistance and capacitance of the LCL lter 1 , 0.05 mF
Pt Total input power of the system 100 kW
T T CDC DC-link capacitance 1.5 mF
Y U H8 UY
us Grid phase-phase voltage 480 V
T T Grid angular frequency 314 rad/s
Y U H9 UY uDCref DC-link voltage reference 900 V
i2qref Q-component reference of the grid-side 0A
Y T U T H10 UY
inductor current
K1p , 1i PI controller parameters of voltage control 0.01, 0.0033
loop
fswitch Switching frequency of the inverter 10 kHz
From the above analysis, the equivalent set of the system Eq.
(18) can be rewritten as

10
According to xi (t) = uij yj (t), the second order approximate

10 
10
j=1
i
y i = i yi + Cpq yp yq + (19) response becomes:
p=1 q=1  10 10 

10 
xi (t) = uij yj0 j
h2kl yk0 yl0 ej t


10 j=1 k=1 l=1 R 
(k,l,j)/
where, y i is the i-th element of Y , Cpq
i = 1
vTik [U T Hk U].  10 10 
2

10 
h2kl j yk0 yl0 e(k +l )t
k=7
+ uij
Considering the initial condition Y0 = [y10 , y20 , ,
y100 ]T , the solution of Eq. (19) can be decomposed into j=1 k=1 l=1 R 
(k,l,j)/
yj (t) = y1j (t)+ y2j (t) + y3j (t) + in the McLaughlin expansion domain  10 10 

10

of convergence. Here, ymj (t) contains the terms that depend on + uij Ckl j yk0 yl0 tej t (22)
any m-states multiples of initial conditions. If the system state
j=1 k=1 l=1 (k,l,j) R
variables are considered as a signal, we can recognize that the
signal is constituted by a number of basic signals. From the Eq. Ckl j
where h2kl j = k +l j
, j, k, l = 1, 2, , 10.
(19), then we can get
 10
The term uij yj0 ej t in the second order approximate response
y 1j = j y1j j=1
is the general expression of linear modal analysis. Therefore, the

10 
10
second modal series method, which gives an explicit correction
y 2j = j y2j + Ckl j y1k y1l (20) term to the linear approximate solution, provides a more accurate
k=1 l=1 approximation of the system than the linear modal method.
..
.
3. Transient phenomena

In this section, we will present a series of simulations to study

Eq. (20) can be solved by one and two dimensions Laplace trans-
the LVRT transient behavior of the grid-connected converter with
form. Neglecting y3j (t) and the other higher order terms, the inverse
LCL lter. All the components of the simulation model are built with
Laplace transform of yj can then be written [24]:
the standard electrical component blocks from the SimPowerSys-
tems block set in Matlab/Simulink library. The parameters used in
the simulation are given in Table 1.

10 
10
The transient dynamic response of the system output current,
yj (t) = y1j (t) + y2j (t) = yj0 ei t + Ckl j yk0 yl0 Skl j (t) = yj0 ei t
the control system reference current and the DC-link voltage are
k=1 l=1 essential for assessing the operation of the system in LVRT pro-
 10 10 
 e(k +l )t ej t
cess. The results of both the second order modal and the linear
j modal approximation simulations, compared to the original non-
+ Ckl yk0 yl0
k + l j linear time-domain simulation results of the system ride-through
k=1 l=1 R 
(k,l,j)/
 10 10  process when an event of three-phase voltage sags to 75% of the
 transmission grid voltage occurred in 0.2 s and ended in 0.7 s, are
+ Ckl j yk0 yl0 tej t (21) shown in Fig. 5. The results show that the modal series method
k=1 l=1 (k,l,j) R can manifest the system behavior more accurately and capture
the dynamic response characteristics more effectively compared
with the linear modal method, especially for the transient process.
where j = 1, 2, , 10. The set R cause the second order resonance Therefore, these simulation results demonstrate that the nonlinear
condition, i.e., satises the condition k + l = j . modal series method is appropriate for transient behavior analysis.
44 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950

Fig. 5. Second order modal series, linear and nonlinear simulation results: (a) Waveform of phase A of L2 current i2a ; (b) waveform of d-component of L2 reference Current
i2dref ; (c) waveform of DC-link voltage uDC .

4. Transient characteristics analysis
In the foregoing section, we have observed that the second order
modal response has maintained the transient operation trend of the
system state variables in the nonlinear time domain simulation.
In this section, we will give a formal theoretical analysis for the
transient phenomena.
4.1. Eigenvalue analysis
Using the Jacobian matrix derived in Section 2, the eigenvalues
of the system can be calculated by the standard method. In the sys-
tem LVRT process, both the voltage dip and voltage recovery states
have 10 eigenvalues that belong to oscillatory modes. The com-
plete system modes are listed in Table 2 during voltage dip and
Table 3 during voltage recovery, respectively. It can be clearly seen
that the system eventually operates at a stable state in the whole
LVRT process. In Tables 2 and 3, 9 ,10 are related to PI controllers
of the grid-connected converter and the DC-link voltage, 1 ,2 and
3 ,4 have a high frequency, 7 ,8 and 9 ,10 have a high damping
ratio, and 5 ,6 have a low degree of participation for the corre-
sponding state variables. For ud and uq , the modes corresponding
to the state variables are signicantly increased in their own par-
ticipation and the damping ratio are increased at the same time
which determines the decay rate of the oscillation amplitude and
the attenuation characteristics. For uod and uoq , the damping ratio
is decreased.
In power systems, the analysis and design of the system usu-
ally follows from a linear systems analysis under neglecting the
possible nonlinear interaction between modes. In fact, a disturb-
ance can excite numerous oscillation modes in the system LVRT
transient process, only a few of these modes are of primary inter-
est to the system designer. The system modes distribution of
both the linear method and the nonlinear second order modal
series method in voltage dip state and voltage recovery state are
shown in Fig. 6 below, respectively. Since the modes distribution is

Table 2
Basic modes analysis during voltage dip.

State variables Participation factor Eigenvalue/103 Frequency/kHz Damping ratio/%

i1d , i1q 0.1265, 0.1253 0.5998 5.2795i 0.8403 0.11288

i2d , i2q 0.1253, 0.0404 0.5973 4.6515i 0.7403 0.12736
uod , uoq 0.000224, 0.0001 0.0602 0.3635i 0.05785 0.16339
ud , uq 0.4326, 0.0170 0.0285 0.0559i 0.008897 0.45421
i2dref , uDC 0.5311, 0.0301 0.0040 0.0002i 0.000032 0.99875
 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950 45

Table 3
Basic modes analysis during voltage recovery.

State variables Participation factor Eigenvalue/103 Frequency/kHz Damping ratio/%

i1d , i1q 0.1270, 0.1248 0.5995 5.2804i 0.8404 0.11281

i2d , i2q 0.1247, 0.1270 0.5976 4.6526i 0.7405 0.12740
uod , uoq 0.000473, 0.000039 0.0347 0.3872i 0.06162 0.08926
ud , uq 0.5021, 0.6069 0.0473 0.0469i 0.007464 0.71010
i2dref , uDC 0.5248, 0.0269 0.0040 0.0002i 0.000032 0.99875

Fig. 6. (a) Linear modes distribution and second order modes distribution during grid voltage dip; (b) Linear modes distribution and second order modes distribution during
grid voltage recovery.

conjugate-symmetric, a zoom-in sub-gure for half parts of the dis- simulation illustrate that the nonlinear system response contains
tribution space is added to show the difference between the linear a number of nonlinear interactive modes compared with the lin-
method and the nonlinear second order modal series method in ear modal simulation. In the nonlinear system response, except for
more detail. It is clear that the modes in the modal series method the foregoing 10 detectable eigenvalues from linear model, much
are much more than in the linear method, as well as the distribu- more interaction modes, like i,j = i + j , also exist. Some of the
tion range is wider than in the linear method. According to Eq. (22), results are shown in Figs. 7 and 8. Based on the modal interac-
the coupling degree of the system will increase due to system has tion indicator I2ikl , these gures show the effects of k,l on the state
a large number interactive mode. Moreover, we can nd out that variables. Where k=5,6,7,8, l=9,10 and k=9,10, l=5,6,7,8 are the dominant
some modes of the system move to the imaginary axis and the oscil- nonlinear interactive modes for the state variable i2d , k=7,8, l=9,10
lation frequency increases in the LVRT process. Therefore, power and k=9,10, l=7,8 the dominant nonlinear interactive modes for the
systems exhibit a complicated dynamic behavior when subjected state variable i2q , i2dref and uDC . The dominant nonlinear interac-
to disturbances, and a linear theoretical analysis of this behavior is tive modes will play a major role in the system transient behavior
not feasible in the grid-connected converter with LCL lter. of the process. According to the denition of I2ikl , the amplitude
and the duration of the transient behavior will be lager and longer
4.2. Nonlinear interaction between two modals when the indicator value of these dominant nonlinear interactive
modes is greater. Here, k=5,6,7,8, l=9 and k=9, l=5,6,7,8 are correspon-
Using Eq. (22), the second order modal time response term dence to the interaction between the AC circuit and the control

10
j
system, k=5,6,7,8, l=10 and k=10, l=5,6,7,8 to the interaction between
uij h2kl yk0 yl0 e(k +l )t is determined by two parameters of the
j=1

10
j
absolute value of the constant uij h2kl yk0 yl0 and the time con-
j=1
stant  kl = 1/real(k + l ). Therefore, I2ikl is dened as an indicator
for the effects of the modal interaction in the time response to cap-
ture both the amplitude and the duration effects of the interactive
modes [24] and expressed as
 10 
 
 
 uij h2kl yk0 yl0 
j
 
 j=1 
I2ikl =   (23)
 real(k + l ) 
 
 
 
It can be seen from the established equation that indicators
can quantify the strength of each modal interaction and provides
Fig. 7. Effects of modal interaction on the d-component of the L2 current i2d , the
a convenient for investigating the dominant nonlinear interactive q-component of the L2 current i2q , the d-component of the L2 reference current i2dref
modes. In addition, the results of the second order modal series and the DC-Link voltage uDC during grid voltage dip.
46 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950

grid-side inductance L2 become larger, respectively. Also shown

Fig. 8. Effects of modal interaction on the d-component of the L2 current i2d , the
q-component of the L2 current i2q , the d-component of the L2 reference current i2dref
and the DC-Link voltage uDC during grid voltage recovery.
the AC circuit and the DC-link. These stronger modal interactions
enhance the coupling degree of the system which will lead to the
analysis of system transient process more difcult and therefore,
the nonlinear modal interaction between the DC side and AC side
of the system, as well as the main circuit and control system, makes
the system LVRT transient dynamic behavior complicated. In addi-
tion, two period analysis of the entire process also shows that the
modal interaction between the grid current and the DC-link voltage
are weakened when the grid voltage turning to recovery.
Since most of the modal interactions are not particularly evi-
dent in engineering practice, we can get the approximate describing
function by selecting the main modal of interaction which is suit-
able for the main state variable. Here, we give the expression of the
main state variables of the system as

10

10 

10 10
in Fig. 10(ad) are the plots of the modal interaction indices vari-
ation of the system state variable in voltage recovery period. It
can be seen from Figs. 9(a and b) and 10(a and b) that the modal
interaction decreases as C and CDC increases. The increases of L1
and L2 result in an increase in the modal interaction as shown in
Figs. 9(c and d) and 10(c and d).
Additionally, it can be observed that the system modal interac-
tions differ greatly in their sensitivities to different parameters and
different operating period. The system parameters CDC , L1 and L2
have the stronger effects on the modal interactions in voltage recov-
ery period, but the effect of C is very little. This observation implies
that the system will get a steady state as quickly as possible by
adjusting the parameters CDC , L1 and L2 . Moreover, the interaction
sensitivities of parameters are also changed in the LVRT process.
In summary, some practical results can be reached for the non-
linear modal interactions:
1) Decreasing L1 and L2 or increasing C and CDC is a feasible method
to reduce the modal interaction. For transient process of specic
state variables, we should consider the modal interactions in
detail.
2) Ignoring the feeble effect of parameter C, it will be able to quickly
increase or decrease the modal interaction by adjusting the
parameters CDC , L1 and L2 during voltage recovery.
3) Opposite the voltage dip period, we should increase the regu-
lated degree of the parameters L1 and L2 which helps the system
to achieve convergence in voltage recovery period.
4.3. Dominant oscillation mode
The dominant oscillation mode is of the greatest impact on both
the system dynamics and the stability of oscillation modal. In lin-
ear theory, the stability of the system is independently dominated

n

j t j j t by every single modal, the response xi (t) = uij yj0 ej t (i = 1, 2,

xi=3 (t) = uij yj0 e uij h2kl yk0 yl0 e
j=1
j=1 j=1 k=1 l=1
 , n) manifests the effects of each modes for the state variable xi ,

10 
8

10 uij uj0  the maximum amplitude of the jth mode at the zero-time,
+ uij h2kl j yk0 yl0 e(k +l )t and Tj = 1/real(j ) the attenuation
 speed of the mode. So the jth
j=1 k=5 l=9 mode, which is of max(uij  /real(j )), is called the dominant oscil-
lation mode due to its biggest inuence on the system dynamics and

10 
10

8
stability. However, the dominant oscillation mode in nonlinear sys-
+ uij h2kl j yk0 yl0 e(k +l )t (24) tem may be different from that in linear system since the nonlinear
j=1 k=9 l=5 oscillation mode is related not only to the fundamental modes, but
also the nonlinear interaction between the modals. Since the loca-

10
 
10 10 10 tion, the type and the duration of the disturbance may be different,
xi=4,9,10 (t) = uij yj0 ej t uij h2kl j yk0 yl0 ej t the motivation level of each oscillation modal will also be different.
j=1 j=1 k=1 l=1
The corresponding strongest incentive and nonlinear interaction
modal is the dominant oscillation mode. Developing the denition

10 
8

10 of the linear dominant oscillation mode into the nonlinear system
+ uij h2kl j yk0 yl0 e(k +l )t analysis theory, the indices of the dominant modal considered in
j=1 k=7 l=9 the nonlinear modal interaction would be dened by analyzing the
second-order analytical solution of the system response. Therefore,

10 
10

8
according to Eq. (22), I(j) is dened as a measure for the response of
+ uij h2kl j yk0 yl0 e(k +l )t (25)
the jth mode and the interaction between the jth mode and other
j=1 k=9 l=7 modes, and we obtain
 
It has been stated in the preceding discussion that the LVRT  maxuji h2j yk0 yl0 maxuji Ckl yk0 yl0 maxuji h2kl yk0 yl0 
j j
 k,l kl
I(j) =  
k,l k,l
transient characteristics is mainly related to the modal interac- +2e1 +
real(j ) real(j ) real(i + j ) 
tions. From a practical standpoint, it is necessary to acquire a  
better understanding of the relation between the modal interac- (27)
tion and the system parameters to make transient process shorter.
Fig. 9(ad) shows the modal interaction indices variation of the
system state variable during voltage dip when DC-link capaci- The maximum modulus value of I(j) indicates that j has the
tance CDC , lter capacitance C, inverter-side inductance L1 and strongest nonlinear interaction, the corresponding modal will be
 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950 47

Fig. 9. Nonlinear interaction variation during voltage dip: (a) The CDC increasing from 1300 F to 2000 F; (b) The C increasing from 45 F to 59 F; (c) The L1 increasing
from 1300 H to 2000 H; (d) The L2 increasing from 1300 H to 2000 H.

Fig. 10. Nonlinear interaction variation during voltage recovery: (a) The CDC increasing from 1300 F to 2000 F; (b) The C increasing from 45 F to 59 F; (c) The L1 increasing
from 1300 H to 2000 H; (d) The L2 increasing from 1300 H to 2000 H.
48 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950

Fig. 11. Dominant oscillation mode of each state variable.

considered as the dominant oscillation mode. This quantitative
approach makes the discrimination of dominant oscillation mode
easier.
We use the index I(j) to denote the jth oscillation mode of each
state variable in the time response and plot it in three-dimensional
coordinates. Since the system experiences a voltage sag and recov-
ery period in the process of LVRT, there may be a certain change for
dominant oscillation mode and also the signicance correspond-
ing to the state variables. As shown in Fig. 11(a and b), X and Y
coordinates are used to denote the eigenvalue and the state vari-
able, and Z is used to show I(j), respectively. It is clearly seen that
the effects of 5 and 6 are the dominant oscillation mode for the
state variables i1d , i1q , i2d and i2q in the whole process. For uod and
uoq , the corresponding mode 7 is the dominant oscillation mode
during voltage dip and 10 is the dominant oscillation mode dur-
ing voltage recovery. 7 is also relevant to the dominant oscillation
mode for ud , uq , i2dref and uDC in the whole LVRT process. In addi-
tion, we nd out that the conspicuousness of the system dominant
oscillation mode has changed throughout the process. The signif-
icance of the dominant oscillation mode of the state variables i1d ,
i1q , i2d and i2q is enhanced obviously, and some other dominant
oscillation modes are signicantly weakened, especially for the
i2dref , and uDC .
4.4. Second order quasi-resonance boundary
Eq. (22) represents the second order resonance responses
of the system model in domain R . When the term

n 
n

n
uij Ckl j yk0 yl0 tej t is added to the linear modal approx-
j=1 k=1 l=1
imate solution, it will increase both the amplitude and the
duration of the nonlinear simulation result with respect to
the nonlinear modal simulation in which resonances have not
occur. Therefore, it is often possible that the system resonance
occurs when the relation k + l = j is satised. Lets call the
requirement |k + l j | |j | the condition of the second order
quasi-resonance and denote the set of all three tuples (k, l, j) by
R . Here, the coefcient must be a much smaller constant. This
quasi-resonance will produce an exponentially decaying sinusoid
signal with a frequency of inter-area oscillation.

Fig. 12. Normal transient phenomenon: (a) waveform of phase A of the L2 current i2a ; (b) waveform of d-component of the L2 reference current i2dref ; (c) waveform of the
DC-link voltage uDC . The second order resonance transient phenomenon; (d) waveform of phase A of the L2 current i2a ; (e) waveform of d-component of the L2 reference
current i2dref ; (f) waveform of the DC-link voltage uDC .
 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950
Fig. 13. Second order quasi-resonance boundary in the parameter space of K2p versus  i : (a) DC-link capacitance CDC ; (b) capacitance of LCL lter C; (c) inverter-side inductance
L1 ; (d) gird-side inductance L2 .
Fig. 12(ac) shows the time-domain simulation waveforms of L2
current of phase A i2a , d-component reference of the current i2dref
in L2 , and the DC-link voltage uDC when the control parameters are
set to K2p = 0.5 and i = 0.15. The duration of the transient process
is seen to be very short, and the overshoot relatively small. But
when the control parameter is above a critical value, the resonance
behaviors occur in transient process. Fig. 12(df) shows the corre-
sponding time-domain waveforms resulting from a step increase
of 2 in K2p . It is shown that the transient responses of i2a , i2dref and
uDC are no longer vanished quickly and the amplitude of the tran-
sient response is much bigger. The preceding analysis about the
essence of such behavior is expounded from a second order reso-
nance perspective, and the simulated transient phenomenons also
match the earlier resonance analysis. The simulation results illus-
trate that the second order resonance has a great inuence on the
transient process.
Usually, we always want the system to operate in a fast tran-
sient over-process and restrain resonance. Hence, how to dene
the boundary of the second order resonance is the principal focus
of our study to avoid resonance in practical applications. From the
previous analysis, we consider that the system resonance occurs
when the parameters match the condition |k + l j |0.0005|j |.
Therefore, the system quasi-resonance boundaries corresponding
to resonance are given to delineate the scope of the practical param-
eter value for ensuring system operation in the normal state. These
boundaries can be readily obtained by using the analytical means
described above. Fig. 13 laid out the system parameters boundaries
in the parameter space of K2p ,  2i , and clearly illustrates the effect of
the main system parameters on the second order quasi-resonance
in the space of K2p  i . In Fig. 13, the boundaries divide the param-
eter space into three parts in which one is the normal operation
region and other two are the second order resonance regions.
The system works in the stable operation when the controller
49
parameters are located in the normal operation region. Otherwise,
if the parameters cross the boundary and enter into the second
order resonance region, both the amplitude and the duration of
transient process will persist larger and longer due to the resonance
occurring. Therefore, the selection of the control parameters should
be always located in the normal operation region so that each state
variable will quickly convergence due to avoiding the occurrence of
second order resonance in the LVRT process. Additionally, the val-
ues of L1 and L2 cannot be made too small since these two second
order quasi-resonance boundary of the parameters will intersect
so that there is only the resonance region.
5. Conclusion
This paper studies the LVRT transient behavior of a grid-
connected converter with LCL-lter using nonlinear modal series
method so as to transfer the transient characteristics research from
the qualitative analysis to quantitative analysis. The obtained sim-
ulation results indicate that the nonlinear modal series method
can manifest the system behavior more accurately and capture
the dynamic response characteristics more effectively. It have been
shown that except for the fundamental modes, the LVRT transient
behavior of the system is signicantly inuenced by the inter-
active modes, particularly by the stronger second order modal
interaction which will make higher coupling degree and exacerbate
transient behavior. Subsequently, the more appropriate functions
which include the higher order nonlinear terms can be obtained
by selecting the more strong interactive modes. On the other hand,
the results also indicate that the oscillation amplitude and duration
of the LVRT transient process can be reduced by adjusting certain
of the parameters so as to weaken the second order modal inter-
action. Additionally, the dominant oscillation mode of the system
is changed in the LVRT process. Finally, the relationships between
50 W. Wang, X. Ma / Electric Power Systems Research 105 (2013) 3950
the second order quasi-resonance boundary and the parameters
are given, which can be used to perform the analysis of resonance
and to design the controllers for the grid-connected converter.
References
[1] M. Liserre, R. Teodorescu, F. Blaabjerg, Stability of photovoltaic and wind tur-
bine grid-connected inverters for a large set of grid impedance values, IEEE
Transactions on Power Electronics 21 (2006) 263272.
[2] Y.M. Chen, Y.C. Liu, S.C. Hung, C.S. Cheng, Multi-input inverter for grid-
connected hybrid pv/wind power system, IEEE Transactions on Power
Electronics 22 (2007) 10701077.
G. Majic,
[3] B. Terzic, A. Slutej, Stability analysis of three-phase PWM converter
with LCL lter by means of nonlinear model, Automatika 51 (2010) 221232.
[4] J.E.G. Carrasco, J.M. Tena, D. Ugena, J. Alonso-Martinez, D. Santos-Martin, S.
Arnaltes, Testing low voltage ride through capabilities of solar inverters, Elec-
tric Power Systems Research 96 (2013) 111118.
[5] S.M. Muyeen, R. Takahashi, J. Tamura, Electrolyzer switching strategy for hydro-
gen generation from variable speed wind generator, Electric Power Systems
Research 81 (2011) 11711179.
[6] M. Garca-Gracia, M.P. Comech, J. Salln, D. Lpez-Anda, O. Alonso, Voltage [21] S.M. Muyeen, R. Takahashi, T. Murata, J. Tamura, Transient stability enhance-
dip generator for wind energy systems up to 5 MW, Applied Energy 86 (2009)
565574.
[7] A.H.M.A. Rahim, E.P. Nowicki, Performance of a grid-connected wind gener-
ation system with a robust susceptance controller, Electric Power Systems
Research 81 (2011) 149157.
[8] F. Iov, A.D. Hansen, P. Srensen, N.A. Cutululis, Mapping of Grid Faults and Grid
Codes, Ris National Laboratory, Technical University of Denmark, Roskilde,
2007.
[9] F. Tang, X.M. Jin, X. Zhou, X.Z. Wu, L. Ma, A novel LVRT strategy for direct-
drive wind turbines based on permanent magnet synchronous generator, in:
Proceedings of IEEE 7th International Power Electronics and Motion Control
Conference, Harbin, China, 2012, pp. 22282233.
[10] S.M. Muyeen, R. Takahashi, M.H. Ali, T. Murata, J. Tamura, Transient stabil-
ity augmentation of power system including wind farms by using ECS, IEEE
Transactions on Power Systems 23 (2008) 11791187.
[11] C.H. Benz, W.T. Franke, F.W. Fuchs, Low voltage ride through capability of a 5 kW
grid-tied solar inverter, in: Proceedings of 14th International Power Electronics
and Motion Control Conference, Ohrid, Macedonia, 2010, pp. 1320.
[12] K.H. Kim, Y.C. Jeung, D.C. Lee, H.G. Kim, LVRT scheme of PMSG wind power sys-
tems based on feedback linearization, IEEE Transactions on Power Electronics
27 (2012) 23762384.
[13] M. Abbes, J. Belhadj, New control method of a robust NPC converter for renew-
able energy sources grid connection, Electric Power Systems Research 88
(2012) 5263.
[14] I.D. Margaris, A.D. Hansen, P. Srensen, N.D. Hatziargyriou, Dynamic security
issues in autonomous power systems with increasing wind power penetration,
Electric Power Systems Research 81 (2011) 880887.
[15] J.F. Conroy, R. Watson, Low-voltage ride-through of a full converter wind tur-
bine with permanent magnet generator, IET Renewable Power Generation 1
(2007) 182189.
[16] C. Saniter, J. Janning, Test bench for grid code simulations for multi-MW wind
turbines, design and control, IEEE Transactions on Power Electronics 23 (2008)
17071715.
[17] A. Mullane, G. Lightbody, R. Yacamini, Wind-turbine fault ride-through
enhancement, IEEE Transactions on Power Systems 20 (2005) 19291937.
[18] T. Odun-Ayo, M.L. Crow, Structure-preserved power system transient stabil-
ity using stochastic energy functions, IEEE Transactions on Power Systems 27
(2012) 14501458.
[19] M. Ouhrouche, Transient analysis of a grid connected wind driven induction
generator using a real-time simulation platform, Renewable Energy 34 (2009)
801806.
[20] C. Jauch, Transient and dynamic control of a variable speed wind turbine with
synchronous generator, Wind Energy 10 (2007) 247269.
ment of variable speed wind turbine driven PMSG with rectier-boost
converter-inverter, in: Proceedings of The International Conference on Elec-
trical Machine, Vilamoura, Portugal, 2008, pp. 400405.
[22] X. Jiao, Q. Gao, Integrated dynamic simulation of photovoltaic installations and
its low voltage ride-through capability analysis, in: Proceedings of Australasian
Universities Power Engineering Conference, Bali, Indonesia, 2012.
[23] H. Geng, D. Xu, Stability analysis and improvements for variable-speed multi-
pole permanent magnet synchronous generator-based wind energy conversion
system, IEEE Transactions on Sustainable Energy 2 (2011) 459467.
[24] N. Pariz, H.M. Shanechi, E. Vaahedi, Explaining and validating stressed power
systems behavior using modal series, IEEE Transactions on Power Systems 18
(2003) 778785.
[25] J. Castell, J.M. Esp, R. Garca-Gil, Guidelines for the design and implementa-
tion of the ARPCC control applied to grid-connected converters, Electric Power
Systems Research 95 (2013) 284291.
[26] E. Twining, D.G. Holmes, Grid current regulation of a three-phase voltage source
inverter with an LCL input lter, IEEE Transactions on Power Electronics 18
(2003) 888895.
[27] M. Chinchilla, S. Arnaltes, J.C. Burgos, Control of permanent-magnet generators
applied to variable-speed wind-energy systems connected to the grid, IEEE
Transactions on Energy Conversion 21 (2006) 130135.