第 36 卷第 12 期 系统仿真学报©

系统仿真学报 Vol. 36 No. 12

2024 年 12 月 Journal of System Simulation Dec. 2024

Harmonic Impedance Modeling and Oscillation Analysis of

Modular Multilevel Converter
Wang Yuhong1，Chen Wensheng1，Gao Shilin1*，Liao Jianquan1，Cheng Yangfan2
(1. College of Electrical Engineering, Sichuan University, Chengdu 610065, China;
2. State Grid Jiaxing Electric Power Supply Company, Jiaxing 314000, China)

Abstract: To facilitate rapid analysis of the oscillation stability mechanism in modular multilevel
converter-based high voltage direct current (MMC-HVDC) systems and streamline the simulation process
for determining MMC impedance characteristics, a simplified mathematical simulation model for MMC
closed-loop impedance is developed using the harmonic state space method. This model considers
various control strategies and includes both AC-side and DC-side impedance models. By applying a
Nyquist criterion-based impedance analysis method, the stability mechanisms on the AC and DC sides of
the MMC are examined. In addition, a data-driven oscillation stability analysis method is also proposed,
leveraging a global sensitivity algorithm based on fast model results to identify key parameters
influencing MMC oscillation stability. Based on sensitivity analysis results, a parameter adjustment
strategy for oscillation suppression is proposed. The simulation results from the MATLAB/Simulink-
based MMC model validate the effectiveness of the proposed method.
Keywords: modular multilevel converter; impedance modeling; harmonic state space; global sensitivity;
impedance analysis

模块化多电平换流器谐波阻抗建模与振荡分析
王渝红 1，陈文晟 1，高仕林 1*，廖建权 1，程杨帆 2
(1. 四川大学 电气工程学院，四川 成都 610065；2. 国家电网嘉兴供电公司，浙江 嘉兴 314000)

摘要：
摘要 ：为快速分析基于模块化多电平换流器的高压直流输电系统(MMC-HVDC)的振荡稳定机理，简
化 MMC 阻抗特性获取过程中的仿真流程，利用谐波状态空间方法建立了考虑不同控制方式的 MMC
闭环阻抗数学简化仿真模型，包含交流侧阻抗模型和直流侧阻抗模型；根据基于奈奎斯特判据的阻抗
分析法研究了MMC交、直流侧的稳定机理；提出了一种基于数据驱动的振荡稳定性分析方法，即基
于快速模型的结果利用全局灵敏度算法分析影响 MMC 振荡稳定的关键参数，并根据灵敏度分析结果，
提出振荡抑制的参数调整策略。在MATLAB/Simulink中建立的MMC模型验证了方法的有效性。
关键词：
关键词：模块化多电平换流器；阻抗建模；谐波状态空间；全局灵敏度；阻抗分析法
中图分类号：TP391 文献标志码：A 文章编号：1004-731X(2024)12-2755-16
DOI: 10.16182/j.issn1004731x.joss.24-FZ0803E
引用格式 王渝红, 陈文晟, 高仕林, 等. 模块化多电平换流器谐波阻抗建模与振荡分析[J]. 系统仿真学报, 2024, 36
引用格式:
(12): 2755-2770.
Reference format: Wang Yuhong, Chen Wensheng, Gao Shilin, et al. Harmonic Impedance Modeling and Oscillation
Analysis of Modular Multilevel Converter[J]. Journal of System Simulation, 2024, 36(12): 2755-2770.

Received date：2024-07-22 Revised date：2024-09-27

Foundation：National Natural Science Foundation of China (52307127); State Key Laboratory of Power System Operation and Control (SKLD23KZ07)
First author：Wang Yuhong(1971-), female, professor, PhD, research area: power system stability and control.
Corresponding author：Gao Shilin(1994-), male, associate researcher, PhD, research area: power system analysis.
第 36 卷第 12 期 系统仿真学报 Vol. 36 No. 12
2024 年 12 月 Journal of System Simulation Dec. 2024

0 Introduction state variable to this small signal disturbance,

linearizing the model, and calculating system
Renewable energy sources are often located far
impedance from the resulting voltage and current.
from power load centers. MMC-HVDC systems
Both methods are based on similar mathematical
facilitate bulk renewable power integration and
principles, but the harmonic state space model offers
transmission, making them widely used in modern
a more convenient representation of state variables
[1-6]
power systems . However, in recent years, oscillation
for matrix operation, making it simpler and more
issues have frequently arisen in MMC-HVDC [14]
efficient for engineering applications . Reference
[7-9]
transmission projects , compromising the safe and
[15] proposed an MMC impedance model based on
stable operation of these systems. In severe cases,
harmonic state space but did not derive the MMC
oscillations can damage system equipment, cause
impedance model under constant power control.
burnout, or even lead to system shutdown. Therefore,
When using detailed simulation models in
it is essential to study the oscillation mechanisms and
Simulink, obtaining simulation results under
develop suppression methods for MMCs.
different parameter settings requires substantial
To better understand the oscillation mechanism
computational resources and time, complicating
and analyze MMC characteristics across various
sensitivity analysis. However, once an accurate
control links, an accurate mathematical model of
mathematical model of MMC impedance is
MMC is necessary. In existing literature, the
established, the factors influencing MMC impedance
equivalent mathematical model of MMC has evolved
characteristics can be further studied. The Sobol
from time-domain to frequency-domain molding and
global sensitivity analysis method allows for a
from single-harmonic to multi-harmonic topologies.
quantitative examination of these factors, helping to
Time-domain models primarily obtain a system’s
identify key parameters relevant to oscillation
state-space model in dq coordinates, calculate the
stability. This method can analyze the effect of
state matrix’s eigenvectors and eigenvalues, and
individual parameters on the system and check for
analyze system stability. However, these models have
interactions between input parameters. All
notable limitations: ① they make it difficult to reveal
parameters can vary simultaneously with different
oscillation mechanisms; ② they have poor scalability;
ranges, and the analysis is not limited by whether the
and ③ changing system operating conditions require
model is linear or additive, making it suitable for
rewriting the state equations, making the solution
application to the MMC impedance model developed
[10-11]
process complex . To overcome these limitations,
in this paper.
MMC impedance analysis methods in the frequency
To accelerate the simulation of MMC
domain have been proposed. Multi-harmonic
impedance characteristics, this paper employs the
topology modeling methods used in the frequency
harmonic state space method to construct a
[12]
domain include multi-harmonic linearization and
mathematically simplified simulation model of
[13]
the harmonic state space method . The basic
MMC, incorporating various control strategies. The
principle involves injecting a disturbance signal at a
results from this model are compared with those
specific frequency, deriving the response of each
from a detailed Simulink model. Unlike the detailed

http: // www.china-simulation.com
• 2756 •
第 36 卷第 12 期 Vol. 36 No. 12
2024 年 12 月 王渝红, 等: 模块化多电平换流器谐波阻抗建模与振荡分析 Dec. 2024

model, the proposed simplified model enables faster where, uZgac, j and uZgdc represent the impedance
analysis of oscillation mechanisms. Using this voltages of the AC and DC sides of the MMC,
simplified model, a data-driven approach for respectively.
oscillation stability analysis is introduced. By rapidly
i u A i dc
generating data with the simplified model and
applying the Sobol global sensitivity algorithm, key SM1 SM1 SM1

parameters influencing oscillation stability are SM2 SM2 SM2

u u j
identified across a broad frequency range. Based on
SMN SMN SMN
these findings, a parameter tuning method for
L arm L arm L arm
oscillation suppression is proposed.
ua R arm R arm R arm
ub
1 Accelerated and Accurate Modeling uc u dc
of MMC Impedance L arm L arm L arm
R arm R arm R arm
1.1 Small-signal Frequency Domain Model u 1 j SM1 SM1 SM1
of MMC
SM2 SM2 SM2
The main topology of the MMC is shown in
SMN SMN SMN
Fig. 1, consisting of six arms. Each arm contains N
i l A
sub-modules along with the corresponding arm
resistance Rarm and arm inductance L arm. In the figure, Fig. 1 Basic structure of MMC

udc and idc represent DC voltage and DC current

Each sub-module in the MMC is composed of a
respectively.
capacitor Cs and a full or half-bridge circuit in
Based on the basic topology of MMC and
parallel with it. To average the voltage and current
Kirchhoff’s law as shown in Fig. 1, the following
across the sub-modules, it is assumed that the
circuit equation can be established. Here, j represents
following equations hold, where mu, j and ml, j are the
a phase, where j∈{A, B, C}.
modulation functions for the upper and lower bridge
1 di u j
u u j = U dc - u j - L arm - R arm i u j (1) arms, respectively:
2 dt
1 di l j Ni Cu j = i u j Nm u j Ni Cl j = i l j Nm l j (5)
u l j = U dc + u j - L arm - R arm i l j (2)
2 dt u u j = u Cu j Nm u j u l j = u Cl j Nm l j (6)
From equations (1) and (2), and considering the To simplify the expressions and capture the
impedance of the AC and DC systems, we obtain: control characteristics of the MMC, differential and
di cm j common mode component forms are defined as
u gdc = 2u cm j + 2R arm i cm j + 2L arm + 3Z gdc i cm
dt
follows:
(3)
x dm j = (x l j - x u j )/2 (7)
u l j -u u j R L d(i l j -i u j )
uj = + arm (i l j -i u j )+ arm + x cm j = (x l j + x u j )/2 (8)
2 2 2 dt
u Zgac j (4)

http: // www.china-simulation.com
• 2757 •
第 36 卷第 12 期 系统仿真学报 Vol. 36 No. 12
2024 年 12 月 Journal of System Simulation Dec. 2024

di cm j 2N N
u gdc = 2u cm j + 2R arm i cm j + 2L arm + 3Z gdc i cm T mcm1 = m DS -1 i cm + m DS -1 i ac + 2Nu Ccm
dt C s cm C s dm
(9) (18)
R arm L di ac j N 2N
u gac j = u dm j + i + arm + Z gac i ac j (10) T mdm1 = m DS -1 i ac + m DS -1 i cm +2Nu Cdm
2 ac j 2 dt C s cm C s dm
du Ccm j 1 (19)
Cs = m cm j i cm j + m dm j i ac j (11)
dt 2 N N
du Cdm j 1 T mcm2 = m DS -1 i cm + m DS -1 i ac + Nu Cdm
Cs = m cm j i ac j + m dm j i cm j (12) C s dm 2C s cm
dt 2
(20)
The basic time-domain model of the MMC is
N N
thus derived. At this stage, the small-signal analysis T mdm2 = m DS -1 i ac + m DS -1 i cm + Nu Ccm
2C s dm C s cm
method is employed to linearize the model, with (21)
Fourier expansion applied to each variable. This where, I is the identity matrix, and ΔS is the
-jθ
approach utilizes Euler’s formulacos θ = 0.5(e + e ) jθ
coefficient matrix for small-signal differential
to decompose the sinusoidal signal into two calculation, containing comprehensive frequency
conjugate positive and negative sequence space information.
vectors with opposite directions. Consequently, each
1.2 MMC Impedance Model Considering
frequency harmonic of the periodic signal is
Control System
expressed in terms of amplitude and phase angle
In the previous section, we derived the
within the frequency domain, where these amplitudes
mathematical model of the MMC main circuit.
and phase angles are constant over time. Through
However, MMCs operate in a closed-loop mode,
these steps, the time-domain model is transformed
with modulation signals generated by the control
into the following frequency-domain constant
system. This section further develops the MMC
mathematical model:

ë Du gac û (
êé Du gdc úù = êé T icm1
ë T icm2
T iac1 úù êé 3Z gdc
+
T iac2 û ë
úù ´
Z gac û ) mathematical model to incorporate the control
system based on the prior derivation. The MMC

êé Di cm úù + êé T mcm1 T mdm1 úù êé Dm cm úù
(13)
control system primarily includes phase-locked loop
ë Di ac û ë T mcm2 T mdm2 û ë Dm dm û (PLL), constant voltage/power outer-loop control,
where： current inner-loop control, and circulating current
2N
T icm1 = 2R arm I + 2L arm DS + m DS -1 m cm + suppression. This paper focuses exclusively on
C s cm
constant voltage and constant power outer-loop
2N
m DS -1 m dm (14)
C s dm controls, as these are predominantly used in practical
N N MMC projects, especially in HVDC and renewable
T iac1 = m DS -1 m dm + m DS -1 m cm (15)
C s cm C s dm
energy grid-tie systems. In these applications, outer-
N N
T icm2 = m dm DS -1 m cm + m DS -1 m dm (16) loop controls employ a combination of constant
Cs C s cm
[15-16]
R L N voltage and constant power methods . This
T iac2 = arm I + arm DS + m DS -1 m dm +
2 2 2C s dm combination directly meets the grids requirement for
N stable voltage and power output, and it is relatively
m DS -1 m cm (17)
2C s cm

http: // www.china-simulation.com
• 2758 •
第 36 卷第 12 期 Vol. 36 No. 12
2024 年 12 月 王渝红, 等: 模块化多电平换流器谐波阻抗建模与振荡分析 Dec. 2024

straightforward to implement and adjust. P ref DU d

Di dref =-G p DP - (25)
It should be noted that the methods used above 1.5U d2
3
to establish the frequency-domain small-signal model DP = (U Di + DU d i d + U q Di q + DU q i q ) (26)
2 d d
are all referred to [15], which also provides the
Du ac Dθ
MMC impedance model under constant voltage G pll
Di dref Dm dmd
control, as follows: DP - Td -
Gp Gi

( )
-1
Di d -
êé Du gdc úù = êé I I úù êé T mcm1
-
T mdm1 úù êé 0 0 úù
Di ac
Td + i d′ + Dm dm
ë Du gac û ë I I û ë T mcm2 T mdm2 û ë G ff21 G ff22 û m′dmd - + m′dmq -
Tq + i q′ +
æ êé T icm1 T i1 úù êé 3Z gdc úù + ö÷
çç + Di q -
çç ë T icm2 T i2 û ë Z gac û ÷÷÷ éê Di cm ùú 0 Gi
Dm dmq
Tq -
çç ÷ (22)
çç éê T mcm1 T mdm1 ùú éê G fb11 0 ùú ÷÷÷ ë Di ac û
çç ê ú ÷ Di cm Dm cm
G fb22 û ÷ø - G
è ë T mcm2 T mdm2 û ë 0 icm

where：
Fig. 2 Structure diagram of constant power control system
ìG fb11 =-G icm
ï
ïG fb22 =-(T d -G iT d + + T q -G iT q + ) P - +
íG =-T G G Gp i dref
ïï ff21 d - i udc + +
ïïG ff22 =[-T d -G i i d′ + - T q -G i i q′ + +(m′dmd - + m′dmq - )]G pll
î P ref
1/(1.5u d )
(23)
where, Gicm is the transfer function of the PR Fig. 3 Structure diagram of constant power outer loop control

controller for circulating current suppression, Gpll is

The form of the MMC model under constant
the closed-loop transfer function of the phase-locked
power outer-loop control is consistent with that in
loop, and Gudc is the transfer function of the voltage
(22), with differences in the coefficient matrix, as
outer-loop. Their specific forms are provided in [15]
shown below:
and are not repeated here.
ìG fb11 =-G icm
This paper further derives the impedance model ï
ïï 3
of the MMC under constant power control, extending ïïG fb22 =-(T d -G iT d + + T q -G iT q + )- G p (u dT d + + u qT q + )
ïï 2
the approach from [15]. First, the structural block ïï G ff21 = 0
ï
diagram of the constant power control system is í G ff22 =[-T d -G i i d′ + - T q -G i i q′ + +(m′dmd - + m′dmq - )+
ï 3
ï G (u i ′ + i u′ + u i ′ + i u′ )]G +
shown in Fig. 2. ï 2 p d d + d d + q q + q q + PLL
ï 3 P
The primary difference between constant power ï G p (i dT d + + i qT q + )+ ref 2 (T d + + u′d +G PLL )
ïî 2 1.5u d
outer-loop control and constant voltage outer-loop
control lies in how the reference value of the d-axis (27)

current is generated, as shown in Fig. 3. where, T d +, T d -, T q +, and T q - represent the frequency

Fig. 3 Structure diagram of constant power shift matrix, with detailed derivation provided in

outer loop control [15]. i d′ +, i d′ -, u′d +, and u′d - represent the steady-state

Thus, the following equations hold: signal’s Toepiltz matrix corresponding to the d-axis

Dm dm =-G i Di q + (Di dref - Di d )G i (24) and q-axis, which can be obtained from the system’s
steady-state operating point.

http: // www.china-simulation.com
• 2759 •
第 36 卷第 12 期 系统仿真学报 Vol. 36 No. 12
2024 年 12 月 Journal of System Simulation Dec. 2024

From this model, we observe that when a impedance of the DC cable. Based on circuit
voltage disturbance signal with a specific frequency principles, the impedance expressions for the two
is applied to the MMC system, the corresponding subsystems are derived as follows:
current response can be calculated, allowing for the Z s1 = Z MMC1 + jωL s (30)
calculation of the MMC impedance. The AC-side Z s2 =jωL dc1 +
impedance of the MMC is: A[Z MMC2 +jω(L s +L dc2 +L dc1 )]+jωL dc2
Du gdc (ω p ) A[1+jωZ MMC2 C dc1 -ω2 C dc1 (L s +L dc2 +L dc1 )]-ω2 L dc2 C dc1
Z AC_MMC (ω p ) = - Z gac (ω p ) (28)
Di ac (ω p ) (31)
The DC side impedance of MMC is: where，
Du gdc (ω p ) A = 1 - ω2 L dc2 C dc2
Z DC_MMC (ω p ) = - Z gdc (ω p ) (29) To assess the resonant stability of the MMC AC
3Di cm (ω p )
side, we compare the equivalent impedance of the
2 AC-side and DC-side Impedance MMC with that of the connected AC system, as
Modeling of MMC-HVDC illustrated by the equivalent circuit shown in Fig. 5.

The schematic diagram of the DC-side

impedance model of an MMC-HVDC system, Lg Rg

including DC cables, is shown in Fig. 4. The

impedance of the MMC-HVDC system is divided MMC


Lg Rg
into two subsystems: subsystem 1 includes the DC-
side impedance and the smoothing inductance Ls of
the DC voltage control station, while subsystem 2 Lg Rg
includes the DC impedance of the MMC on the Fig. 5 Equivalent circuit of MMC connected to AC grid
power control side, the smoothing inductance Ls, and
the impedance of DC cable 1 and DC cable 2. Similarly, the impedance expression for the two
subsystems is as follows:
Ls L dc1 L dc1 Ls Z s1 = R g + jωL g (32)
Z s2 = Z MMC (33)
C1 With the AC/DC-side impedance model of the
Z MMC1
Z MMC2
MMC established, impedance analysis can be applied
to examine system stability. The principle involves
Subsystem 1
C2
dividing the power system into two subsystems,
where each subsystem’s impedance model is
L dc2 L dc2
relatively independent and is typically unaffected by
Equivalent impedance Subsystem 2
of DC line the parameters or structural changes of the other
subsystem. According to the Nyquist stability
Fig. 4 Equivalent impedance diagram at DC-side
criterion, if the impedance ratio of the two systems

A cascade lumped circuit is used to model the has a phase angle margin greater than 0°, the system

http: // www.china-simulation.com
• 2760 •
第 36 卷第 12 期 Vol. 36 No. 12
2024 年 12 月 王渝红, 等: 模块化多电平换流器谐波阻抗建模与振荡分析 Dec. 2024

remains stable. The criterion implies that when the when a single input changes alone; the third term
impedance magnitudes of the two subsystems represents the function when two inputs change
intersect within the analyzed frequency band, and the simultaneously; and so forth. Each decomposition
phase angle difference between their impedances is subterm has a zero integral for any input factor it
[9]
less than 180°, the system is stable . contains:

∫
1

3 Data-driven Oscillation Stability f i1 i2 is (x i1 x i2 x in )dx in = 0 (35)

0

Analysis Method Based on These formulas show that the subterms in (35)
Accelerated Simulation Results are orthogonal, meaning that when (i 1 i 2 i s ) ¹
( j 1 j 2 j s ), the following equation holds:
To quantitatively evaluate the influence of each
factor on MMC impedance, sensitivity values for ∫ Ω
f i1 i2 is × f j1 j2 js dX = 0 (36)
each input parameter can be calculated using relevant where, Ω represents all value ranges of X during
mathematical methods. Parameters with a greater continuous integration.
impact on MMC impedance can then be identified From this theory, to solve f0, we need only
and refined based on these results. integrate X on both sides of (34):
The main principle of the Sobol method is to
decompose the variance of model output. This
f0 = ∫ Ω
f (X )dX (37)

By integrating parameters other than xis on both

approach divides the target model into a function of
sides of (34), we get:
each individual input parameter and combinations of
different input parameters, calculating the ∫ Ω
f (X ) ∏dx n = f 0 + f is (x is )
n ¹ is
(38)
contribution of the parameter’s variance to the total
All terms on the right-hand side of (36) can be
variance. This process measures both the impact of
analogously obtained.
individual parameters on the system and the
According to these conditions, the total variance
interactions between input parameters. The basic
of model f(X) can be obtained as follows:

∫
[17-18]
principle is described below .
D= f 2 (X )dX - f 02 (39)
The model under analysis is defined as Y = f (X ), Ω

with a definition domain in the n-dimensional unit The S-order partial variance is defined as:

∫ ∫
1 1
body, where n represents the number of input D i1 i2 is =  f i1 i2 is (x i1 x i2 x is )
0 0
parameters in the model, X=(x1, x2,…, xn). The Sobol
dx i1 dx i2dx is (40)
method decomposes Y into a sum of individual
The relationship between total variance and
elements in X and combinations of elements, as
partial variance of each order is as follows:
shown below:
D = ∑D i + ∑
n

f (X ) = f 0 + ∑f i (x i ) + ∑
n D i j +  + D 12n (41)
f i j (x i x j ) +  + i=1 1≤i<j≤n
i=1 1≤i<j≤n
The sensitivity coefficient of a certain order is
f 12n (x 1 x 2 x n ) (34)
defined as the ratio of this order’s partial variance to
In this formula, the first term f0 represents a
the total variance, as shown below:
constant; the second term represents the effect function
S i1 i2 is = D i1 i2 is /D (42)
http: // www.china-simulation.com
• 2761 •
第 36 卷第 12 期 系统仿真学报 Vol. 36 No. 12
2024 年 12 月 Journal of System Simulation Dec. 2024

When the Sobol method is used for global The first-order sensitivity coefficient indicates
sensitivity analysis, the integrals in (39) and (40) are the independent influence of a factor on the output,
estimated by Monte Carlo sampling. The Monte while the global sensitivity coefficient reflects its
Carlo simulation method approximates solutions to total contribution to output variance, including
technical problems through experiments and random interactions with other factors. If a large difference
sampling. When the sampling size is large enough, exists between a parameter’s first-order and global
this method is highly applicable. f0 and D are sensitivity coefficients, it indicates significant
obtained through Monte Carlo estimation. interaction with other factors. Conversely, if the first-

N∑
N
1 order and global sensitivity coefficients are similar, it
f0 = (X m ) (43)
m=1 suggests the input factor has a relatively independent
D = ∑f 2 (A)j - f 02
1 N
(44) impact on the system.
N j=1
The first-order sensitivity coefficient Si is 4 Case Studies
defined as follows:
4.1 Validation of Accurate Impedance
N∑
1 N
B )j - f (A)j )
f (B)j ( f (A(i) Model
Di j=1
Si = = (45)
D D
To verify the correctness of the theoretical
The total sensitivity of each parameter, defined
derivation above, an MMC simulation is established
as its global sensitivity ST is as follows:
in Simulink, with the main electrical parameters

2N ∑
1 N
( f (A)j - f (Α(i)
B )j )
2
shown in Table 1.
j=1
ST = (46)
D Table 1 Main electrical parameters of MMC simulation
In (43)~(46), N represents the number of model
Parameter Symbol Value
sampling points, A and B are the original sample
Number of bridge arm submodules N 216
matrices, and Α(i)
B is the sample matrix generated after
Rated voltage of submodule capacitance/kV uCN 1.6
replacing the i-th column of A with that of B. The Capacitance value of submodule/mF C 10
implementation process is as follows: First, define Arm inductance/ mH L0 60

the number of input factors m, and the number of Arm resistance/Ω R0 1

DC side voltage/kV Udc ±320
samples N. Use Monte Carlo or a variant method to
Effective value of AC measured voltage/kV us 220
obtain initial sample matrices AN×m and BN×m with N Equivalent inductance of AC grid/mH Ln 220
rows and m columns. Replace the i-th column of Equivalent resistance of AC grid/Ω Rn 14
matrix A with the i-th column of matrix B, yielding a
In this simulation model, the impedance of the
sample matrix Α(i)
B with N rows and m columns. Thus,
MMC at different frequencies is calculated using a
a total of m+2 matrices are obtained, with a
frequency sweep program. The principle is to add a
combined total of (m+2)×N input data sets. This data
disturbance voltage source at the port, apply a
can be substituted into the established impedance
disturbance signal at a given frequency, and measure
model to calculate each input parameter’s first-order
the current response at that frequency once the
sensitivity coefficients and the global sensitivity
system stabilizes, yielding the system’s equivalent
coefficient for MMC impedance using (44)~(46).

http: // www.china-simulation.com
• 2762 •
第 36 卷第 12 期 Vol. 36 No. 12
2024 年 12 月 王渝红, 等: 模块化多电平换流器谐波阻抗建模与振荡分析 Dec. 2024

impedance at the port. This result is then compared Table 2 Controller parameters of MMC model
Parameter Value
with the MMC impedance calculated from the
PLL parameters kpll_p=0.008, kpll_i=1
mathematical model, as shown below.
inner-loop control parameters ki_p= 45, ki_i= 1 000
First, the open-loop impedance model of the DC voltage control parameters ku_p= 0.1, ku_i=1
MMC derived in Section 1.1 is validated. Using the power control parameters Kp_PQ= 8×10-6, Ki_PQ= 8×10-4;

AC side of the constant power control station as an circulating current suppressor Kpicm= 50, Kricm= 500
parameters/(rad/s) ωr= 200 π, ωc= 10
example, the modulation wave is set as follows:
system delay/μs Tds= 400
m cm = 0.5 + cos(2ωt + 174.5°)
m dm = 0.43cos(ωt - 2.5°) Frequency scanning results
65 Model calculation results
The open-loop impedance at the AC side of the 60

Amplitude/dB
55
MMC across different frequency bands is obtained
50
by sweeping the frequency of the simulation open- 45
40
loop model. The results are then compared with the
35
theoretical calculation curve of the harmonic state 30 0 1 2 3 4
10 10 10 10 10
space model, as shown in Fig. 6. Frequency/Hz
200
Frequency scanning results
60 Model calculation results 100
Phase/(°)

50
Amplitude/dB

40 0
30 -100 Frequency scanning results
20 Model calculation results
10 -200 0 1 2 3 4
10 10 10 10 10
0 0 1 2 3 4
10 10 10 10 10 Frequency/Hz
Frequency/Hz (a) AC side impedance verification
100
Frequency scanning results
60
50 Model calculation results
Phase/(°)

Amplitude/dB

50
0
40
-50 Frequency scanning results
30
Model calculation results
-100 0 1 2 3 4 20
10 10 10 10 10 0 1 2 3 4
10 10 10 10 10
Frequency/Hz
Frequency/Hz
Fig. 6 Open-loop impedance verification diagram of 100
constant power control MMC
Phase/(°)

50

The simulation parameters of the MMC control 0

Frequency scanning results
system are shown in Table 2. Model calculation results
-50 0
The MMC impedance model incorporating the 10 10
1
10
2
10
3
10
4

control system is compared with the frequency sweep Frequency/Hz

simulation results for both DC-side and AC-side (b) DC side impedance verification

impedance under constant voltage and constant Fig. 7 Constant voltage control MCC impedance verification
diagram
power control, as shown in Figs. 7 ~ 8.

http: // www.china-simulation.com
• 2763 •
第 36 卷第 12 期 系统仿真学报 Vol. 36 No. 12
2024 年 12 月 Journal of System Simulation Dec. 2024

70
Frequency scanning results under constant power control, the errors were 2.853%
65 Model calculation results
Amplitude/dB

and 1.031% for the AC and DC sides, all within 3%,

60
55 validating the accuracy of the harmonic state space-
50
based MMC model developed in this paper.
45
[19-20]
40 0 1 2 3 4
Some studies only consider partial MMC
10 10 10 10 10
Frequency/Hz
controllers or reply on simplified models rather than a
150 full control system based on MMC. To demonstrate the
100 necessity of a comprehensive modeling and control
Phase/(°)

50
0 process, a control step was intentionally omitted during
-50 the modeling process, resulting in the output shown in
Frequency scanning results
-100 Model calculation results Fig. 9. As observed in Fig. 9, neglecting any part of the
-150 0 1 2 3 4
10 10 10 10 10 control loop significantly increases the model’s
Frequency/Hz
calculation error in the 0~100 Hz frequency band,
(a) AC side impedance verification
rendering results in this range unreliable. However,
65 Model calculation results
60 Frequency scanning results ignoring the control link has minimal impact on
Amplitude/dB

55
50 frequencies above 500 Hz. Therefore, an accurate
45 impedance mathematical model, including a full control
40
35 system, is essential when analyzing system impedance
30 0
10 10
1
10
2
10
3
10
4
characteristics below 500 Hz.
Frequency/Hz
100 Frequency scanning results
Model calculation results Model calculation results
Frequency scanning results Ignore phase-locked loop
Phase/(°)

50 Ignore circulation suppression

Ignore the current inner loop
0 Ignore the voltage outer loop
80
Magnitude/dB

-50 60
0 1 2 3 4
10 10 10 10 10 40
Frequency/Hz
20
(b) DC side impedance verification
0
Fig. 8 Constant power control MCC impedance verification 0 1 2 3 4
diagram 10 10 10 10 10
Frequency/Hz

Figs. 7~8 show that the simulated frequency 200

sweep results of DC and AC impedance under both 100

Phase/(°)

constant power and constant voltage control are 0

consistent with the theoretical results of the derived -100

model. After data extraction and calculation, the -200

0 1 2 3 4
10 10 10 10 10
average relative errors of impedance amplitudes for the Frequency/Hz
AC and DC side models under constant voltage control Fig. 9 Impact of neglecting some control steps on
were 1.135% and 2.852%, respectively. Similarly, impedance calculation results

http: // www.china-simulation.com
• 2764 •
第 36 卷第 12 期 Vol. 36 No. 12
2024 年 12 月 王渝红, 等: 模块化多电平换流器谐波阻抗建模与振荡分析 Dec. 2024

Next, the derived DC cable impedance Subsystem 2

80 Subsystem 1
calculation formula (31) is verified using DC cable 60

Amplitude/dB
parameters in Table 3. 40
20 531 Hz
0 23 Hz
Table 3 DC cable parameters
−20 4 173 Hz
Parameter DC cable 1 DC cable 2 −40 3 477 Hz
Length/km 10.7 10.7 −60 0 1 2 3 4
10 10 10 10 10
Resistance/(Ω/km) 0.01 0.01 Frequency/Hz
Inductance/(mh/km) 0.198 0.107 100
Capacitance/(μF/km) 0.187 0.239 50

Phase/(°)
23 Hz, 76.5° 531 Hz, 83.2°
0
With the smoothing inductance Ls set to 50 mh,
−50 23 Hz, −8° 531 Hz, −85.3°
Fig. 10 shows the comparison between the DC cable
−100 0
line impedance frequency sweep simulation and the 10 10
1
10
2
10
3
10
4

Frequency/Hz
results from (31), confirming the correctness of the
Fig. 11 DC side impedance Bode diagram of two subsystems
derived formula.

Frequency scanning results It is observed that within the 1~6 000 Hz

Model calculation results
100 frequency range, the phase angle difference between
Amplitude/dB

50 the two subsystems remains below 180° , which

satisfies system stability conditions per the Nyquist
0
criterion. However, at 531 Hz, the impedance
-50 101 10
2
10
3
10
4
amplitudes of the two systems are equal, with a phase
Frequency/Hz
difference of 168.5°. At this point, the phase margin for
100
stability is small, indicating a potential oscillation risk if
50 Frequency scanning results
Phase/(°)

Model calculation results the system introduces additional negative damping.

0
Notably, the impedance curve of subsystem 2
-50
shows pronounced resonance peaks near 3 476 Hz
-100 0 1 2 3 4 and 4 173 Hz, suggesting that oscillations may occur
10 10 10 10 10
Frequency/Hz within subsystem 2. To investigate further, subsystem
Fig. 10 Verification of DC cable line impedance 2 is divided into MMC impedance and DC cable
impedance subsystems, and their impedance curves
4.2 Analysis of System Oscillation Stability are calculated as shown in Fig. 12.
4.2.1 Analysis of DC-side Impedance Stability of The impedance curve shows that the phase
MMC difference between MMC2 and line impedance is

Based on the established DC-side impedance less than 180° across the 1~10 000 Hz range.

model of the MMC, the impedance characteristics of According to the Nyquist criterion, the system

the two subsystems in Fig. 4 across different remains stable. By examining the equivalent

frequencies are shown in Fig. 11. impedance curve of the line, it is inferred that the DC
cable and smoothing inductance form an oscillation

http: // www.china-simulation.com
• 2765 •
第 36 卷第 12 期 系统仿真学报 Vol. 36 No. 12
2024 年 12 月 Journal of System Simulation Dec. 2024

circuit around 493, 3 480, and 4 184 Hz. Equivalent difference meets stability conditions, suggesting that the
circuits resembling parallel resonance appear near system can remain stable near this frequency.
493 and 4 184 Hz, causing impedance to rise, while a
Impedance of MMC
series resonance-like circuit forms near 3 480 Hz, 80 Equivalent impedance of power grid

reducing impedance to zero. Setting the impedance 64 Hz

Magnitude/dB
60 43 Hz
of MMC2 to zero in equation (31) and evaluating at 40
these three frequencies yields results that match this 20
behavior, validating the above analysis. 0 0 1 2 3 4
10 10 10 10 10
Equivalent impedance of DC cable Frequency/Hz
100 Impedance of MMC 200 64 Hz, 164°
4 183 Hz 60 Hz, 88°
493 Hz 100
Amplitude/dB

Phase/(°)
50
0 43 Hz, 87°
0
−100 43 Hz, −176°
−50
3 491 HZ −200 0
−100 0 1 2 3 4 10 10
1
10
2
10
3
10
4

10 10 10 10 10
Frequency/Hz
Frequency/Hz
100 Fig. 13 MMC AC side equivalent impedance and AC grid
impedance
50
Phase/(°)

0
This was validated in the simulation model. By
−50
varying AC grid impedance in the stable operating
−100 0 1 2 3 4 system to simulate this condition, oscillations were
10 10 10 10 10
Frequency/Hz observed, with the AC waveform shown in Fig. 14.
Fig. 12 Bode diagram of MMC impedance and DC cable
impedance subsystem 750
500
From this, it can be concluded that the DC-side 250
iac/A

0
system remains stable within the established model.
−250
4.2.2 Stability Analysis of MMC AC-side −500
−750
According to Fig. 5, assuming AC grid 0 0.2 0.4 0.6 0.8 1.0
impedance Rg is 5 Ω and Lg is 0.22 H, the Baud plot
t/s
Fig. 14 AC oscillation waveform
of the grid impedance and MMC impedance is
calculated and shown in Fig. 13.
The FFT analysis of the waveform in Fig. 14 is
By examining the plot, we observe that at
shown in Fig. 15, showing harmonics concentrated
approximately 43 Hz, the phase difference at the
around 43 Hz, which verifies the theoretical analysis.
intersection of impedance amplitudes of the two
4.3 Key Parameter Selection and Parameter
systems exceeds 180° , indicating instability near this
Tuning for Oscillation Suppression
frequency. However, at 64 Hz, where the impedance
amplitudes of the two systems intersect again, the phase For analysis, the constant voltage control side

http: // www.china-simulation.com
• 2766 •
第 36 卷第 12 期 Vol. 36 No. 12
2024 年 12 月 王渝红, 等: 模块化多电平换流器谐波阻抗建模与振荡分析 Dec. 2024

MMC is used as an example. Factors influencing MMC The impact of these factors is analyzed in two
impedance include controller parameters, main circuit frequency ranges: 0~100 Hz and 100~5 000 Hz.
parameters, and system delay. The PI controller Figs. 16~19 show the first-order and total sensitivity
parameters for both voltage and current loops, as well of AC-side impedance factors across frequencies
as the PLL parameters, are chosen due to their under constant voltage control.
significant role in determining system response to load
Sobol-S
and grid condition changes. These parameters are ku_p ku_i ki_p ki_i kpll_p
kpll_i Kpicm Kricm Td sum
crucial for tuning the system to maintain stability across 1.0

First order sensitivity

different operating scenarios. System delays are also 0.8
0.6
included in the analysis, as they represent a practical
0.4
challenge in MMC operation, affecting real-time
0.2
control strategies due to signal processing and actuation 0
20 40 60 80 100
delays. Since sub-module capacitance and bridge arm
Frequency/Hz
inductance are generally fixed post-design in practical
Fig. 16 First-order Sensitivity of AC-side Impedance in
applications, only controller parameters and system 1~100 Hz
delays are analyzed here. The sampling range for each
Sobol-ST
parameter is defined in Table 4, considering system ku_p ki_p kpll_p Kpicm Tds
1.4 ku_i ki_i kpll_i Kricm sum
stability margins.
Global total sensitivity

1.2
Fundamental (50 Hz)=4 320，THD=9.47% 1.0
100 0.8
0.6
Mag(% of Fundamental)

80 0.4
X 43.333 3 0.2
60 Y 52.52 0
40 20 40 60 80 100
Frequency/Hz
20
Fig. 17 Total Sensitivity of AC-side Impedance in 1~100 Hz
0
0 10 20 30 40 50 60 70 80 90 100 Sobol-S
Frequency/Hz
1.0
First order sensitivity

Fig. 15 FFT analysis results of AC 0.8

ku_p kpll_i
0.6 ku_i Kpicm
Table 4 Input parameters and their value ranges
0.4 ki_p Kricm
Parameter Symbol Value ki_i Td
ku_p 0.05~0.3 0.2 kpll_p sum
Voltage outer loop PI parameters
ku_i 0.5~2.5 0
1 000 2 000 3 000 4 000 5 000
ki_p 30~60 Frequency/Hz
PI parameters of current inner loop
ki_i 700~1 300
Fig. 18 First-order Sensitivity of AC-side Impedance in
kpll_p 0.004~0.016 100~5 000 Hz
PI parameters of PLL
kpll_i 0.5~1.5
Kpicm 30~70 From Figs. 17 and 19, we conclude that in the
Circulation control PR parameter
Kricm 350~650
1~100 Hz range, the primary factors affecting MMC
System delay Tds/μs 0~500

http: // www.china-simulation.com
• 2767 •
第 36 卷第 12 期 系统仿真学报 Vol. 36 No. 12
2024 年 12 月 Journal of System Simulation Dec. 2024

impedance are the proportional parameters ku_p and In Fig. 20, increasing ku_p around 40 Hz raises
ki_p of the voltage outer loop and current inner loop the MMC impedance curve within the 40~100 Hz
controllers. Beyond 85 Hz, ki_p becomes the dominant band with a minima impact on other frequencies. By
factor. In the sub-synchronous range, kpll_p and Kpicm analyzing the cases with ku_p=0.25 and ku_p=0.08, a
also have a certain influence on MMC impedance, comparison of impedance curves was made by
though their impact is relatively low in the 1~100 Hz introducing the AC system impedance curve, as
range. In the high-frequency range of 100~5 000 Hz, shown in Fig. 21.
system delay Tds, ki_p, and kpll_p become the main
60

Amplitude/dB
influencing factors. The sensitivity coefficient of 50
ku_p decreases with increasing frequency, approaching 40 ku_p=0.08
ku_p=0.15
30 ku_p=0.25
zero within the 60~400 Hz range. System delay Tds ku_p=0.35
20
becomes the dominant factor at higher frequencies. 10
0
10
1 2
10 10
3
10
4

Frequency/Hz
Sobol-ST 200
1.2
100
Phase/(°)
Global total sensitivity

1.0 ku_p=0.08
0 ku_p=0.15
0.8
ku_p kpll_i −100 ku_p=0.25
0.6 ku_p=0.35
ku_i Kpicm −200 0 1 2 3 4
0.4 ki_p Kricm 10 10 10 10 10
ki_i Td Frequency/Hz
0.2 kpll_p sum
0 Fig. 20 Impedance curves of MMC under different ku_p
1 000 2 000 3 000 4 000 5 000 values
Frequency/Hz
Fig. 19 Total Sensitivity of AC-side Impedance in 50
68 Hz
Amplitude/dB

100~5 000 Hz 45
40 Grid impedance
ku_p=0.08
35 ku_p=0.15
Observing the first-order and total sensitivity 63 Hz
30 41 Hz
coefficients, we find that in the 0~40 Hz range, the 2
10
significant differences between them indicate strong Frequency/Hz
200
interactions among influencing factors, which should
100
Phase/(°)

be consistent when tuning parameters. However, in 0 Grid impedance

ku_p=0.08
the med-to-high frequency range, the differences are −100 43 Hz 66 Hz ku_p=0.15
smaller, indicating that the effects on the system are −200 0 1 2 3 4
10 10 10 10 10
relatively independent and can be adjusted separately. Frequency/Hz

To address sub-synchronous oscillations on the Fig. 21 MMC impedance curve and grid impedance curve

AC side, ku_p was identified as the dominant factor at

Observing Fig. 21, it can be seen that while the
44 Hz. By adjusting ku_p in the MMC impedance
oscillation point around 40 Hz was avoided by the
model, impedance curves were obtained, showing
upward shift of the MMC impedance curve at ku_p=
that ku_p primarily affects the 10~100 Hz range,
0.25, a new intersection between the MMC
supporting the sensitivity analysis results.

http: // www.china-simulation.com
• 2768 •
第 36 卷第 12 期 Vol. 36 No. 12
2024 年 12 月 王渝红, 等: 模块化多电平换流器谐波阻抗建模与振荡分析 Dec. 2024

impedance curve and the grid impedance curve impedance of the MMC, highlighting a potential
emerged in the 66~100 Hz frequency band. At this stability concern.
point, the phase difference exceeded 180°, indicating (3) Applying the Sobol method, the sensitivity
another potential oscillation point. To address this, coefficients of various parameters across a broad
ku_p was reduced to 0.08, which shifted the frequency range were calculated. Key factors affecting
intersection point of the two impedance curves to the impedance characteristics in the low-frequency band
left, effectively avoiding both the original and new include the proportional parameters of the voltage outer
oscillation points. This adjustment met the stability loop and current inner loop controllers (ku_p and ki_p),
requirements. This simulation model verified this while system delay is the dominant factor in mid-to-
result, and the AC waveform, shown in Fig. 22, high frequency oscillations. Furthermore, parameter
confirmed that the system could operate stably, interactions are stronger in the low-frequency band than
supporting the theoretical analysis. in the high-frequency band. Adjustments based on each
parameter’s sensitivity can effectively mitigate
750
500 oscillation risks.
250
0 References：
：
iac/A

−250 [1] Mai Huong Nguyen, Tapan Kumar Saha, Mehdi Eghbal.
−500 Input/Output Selection for Wide-area Power Oscillation
−750 Damper of Hybrid Multi-terminal High-voltage Direct
0 0.1 0.2 0.3 0.4 0.5 Current to Connect Remotely Located Renewable Energy
t/s Resources[J]. IET Generation, Transmission &
Fig. 22 AC of System at ku_p=0.08 Distribution, 2015, 9(5): 483-493.
[2] Zhao Ligang, Xu Guangyue, Hong Chao, et al. Fast
Simulation of MMC-HVDC Systems Using a Generalized
5 Conclusions
Equivalent Circuit Model of MMC[C]//2020 IEEE 16th
International Conference on Control & Automation (ICCA).
This paper analyzes the oscillation stability of
Piscataway: IEEE, 2020: 1605-1611.
the MMC-HVDC system using the proposed [3] Debnath S, Qin Jiangchao, Bahrani B, et al. Operation,
harmonic state space model of the MMC. The Control, and Applications of the Modular Multilevel
Converter: A Review[J]. IEEE Transactions on Power
following conclusions are drawn.
Electronics, 2015, 30(1): 37-53.
(1) The simplified impedance model of the [4] Jose I Leon, Sergio Vazquez, Leopoldo G Franquelo.
MMC, developed using the harmonic state space Multilevel Converters: Control and Modulation
Techniques for Their Operation and Industrial
method, was validated through detailed simulation,
Applications[J]. Proceedings of the IEEE, 2017, 105(11):
demonstrating high accuracy. This model not only 2066-2081.
streamlines the simulation process for MMC [5] Jaime Peralta, Hani Saad, Sébastien Dennetiere, et al.
Detailed and Averaged Models for a 401-level MMC-
impedance characteristics but also provides a
HVDC System[J]. IEEE Transactions on Power Delivery,
foundation for analyzing system stability and 2012, 27(3): 1501-1508.
sensitivity to influencing factors. [6] Rong Fei, Gong Xichang, Huang Shoudao. A Novel Grid-
connected PV System Based on MMC to Get the
(2) Using impedance analysis, this paper
Maximum Power Under Partial Shading Conditions[J].
identifies an oscillation point in the AC-side IEEE Transactions on Power Electronics, 2017, 32(6):

http: // www.china-simulation.com
• 2769 •
第 36 卷第 12 期 系统仿真学报 Vol. 36 No. 12
2024 年 12 月 Journal of System Simulation Dec. 2024

4320-4333. Magazine, 2000, 20(4): 70-81.

[7] Zhang Haitao, Saeedifard M, Wang Xiuli, et al. System [14] Lü Jing, Zhang Xin, Cai Xu, et al. Harmonic State-space
Harmonic Stability Analysis of Grid-tied Interlinking Based Small-signal Impedance Modeling of a Modular
Converters Operating Under AC Voltage Control Mode Multilevel Converter with Consideration of Internal
[J]. IEEE Transactions on Power Systems, 2022, 37(5): Harmonic Dynamics[J]. IEEE Transactions on Power
4106-4109. Electronics, 2019, 34(3): 2134-2148.
[8] Lü Jing, Cai Xu, Amin M, et al. Sub-synchronous [15] Xu Zigao, Li Binbin, Han Linjie, et al. A Complete HSS-
Oscillation Mechanism and Its Suppression in MMC- based Impedance Model of MMC Considering Grid
based HVDC Connected Wind Farms[J]. IET Generation, Impedance Coupling[J]. IEEE Transactions on Power
Transmission & Distribution, 2018, 12(4): 1021-1029. Electronics, 2020, 35(12): 12929-12948.
[9] Zou Changyue, Rao Hong, Xu Shukai, et al. Analysis of [16] Zhang Yang, Chen Xin, Sun Jian. Sequence Impedance
Resonance Between a VSC-HVDC Converter and the Modeling and Analysis of MMC in Single-star
AC Grid[J]. IEEE Transactions on Power Electronics, Configuration[J]. IEEE Transactions on Power
2018, 33(12): 10157-10168. Electronics, 2020, 35(1): 334-346.
[10] 郭春义, 殷子寒, 赵成勇. MMC-HVDC系统在整流和逆 [17] Sobol′ I M. Global Sensitivity Indices for Nonlinear
变工作模式下的小干扰稳定性对比研究[J]. 中国电机 Mathematical Models and Their Monte Carlo Estimates
工程学报, 2018, 38(24): 7349-7358. [J]. Mathematics and Computers in Simulation, 2001, 55
Guo Chunyi, Yin Zihan, Zhao Chengyong. Comparative (1/3): 271-280.
Study on Small-signal Stability of MMC-HVDC System [18] Zhang Chi, Chu Jinggang, Fu Guangtao. Sobol's
Under Rectification and Inversion Modes[J]. Proceedings Sensitivity Analysis for a Distributed Hydrological
of the CSEE, 2018, 38(24): 7349-7358. Model of Yichun River Basin, China[J]. Journal of
[11] Wang Wenyuan, Barnes M, Marjanovic O, et al. Impact Hydrology, 2013, 480: 58-68.
of DC Breaker Systems on Multiterminal VSC-HVDC [19] Chen Qiang, Lü Jing, Li Rui, et al. Impedance Modeling
Stability[J]. IEEE Transactions on Power Delivery, 2016, of Modular Multilevel Converter Based on Harmonic
31(2): 769-779. State Space[C]//2016 IEEE 17th Workshop on Control
[12] Sun Jian, Liu Hanchao. Sequence Impedance Modeling and Modeling for Power Electronics (COMPEL).
of Modular Multilevel Converters[J]. IEEE Journal of Piscataway: IEEE, 2016: 1-5.
Emerging and Selected Topics in Power Electronics, [20] Xu Zigao, Li Binbin, Wang Shengbo, et al. Generalized
2017, 5(4): 1427-1443. Single-phase Harmonic State Space Modeling of the
[13] Mollerstedt E, Bernhardsson B. Out of Control Because Modular Multilevel Converter with Zero-sequence
of Harmonics-an Analysis of the Harmonic Response of Voltage Compensation[J]. IEEE Transactions on
an Inverter Locomotive[J]. IEEE Control Systems Industrial Electronics, 2019, 66(8): 6416-6426.

http: // www.china-simulation.com
• 2770 •