Department of Energy Technology, Aalborg University, Denmark

Modular Multi-Level Converter:

Modeling, Simulation and Control
in Steady State and Dynamic
Conditions

1
2
Title: Modular Multi-Level Converter: Modeling, Simulation and
Control in Steady State and Dynamic Conditions
Semester: WPS2 (10th)
Semester theme: []
Project period: 01/02/12 to 30/05/12
ECTS: 30
Supervisor: Remus Teodorescu
Luca Zarri

SYNOPSIS:

The aim of this project is the analysis

of a Modular Multilevel Converter
(MMC) and the development of a
control scheme for energy stored.
Project group: [WPS2 1056] The analysis is based on the use of a
simplified circuit constituted by a
_____________________________________
single leg of the converter where all
[Name 1]
the modules in each arm are
represented by a single variable
voltage source. Control scheme are
thus investigated, focusing on the
balancing between upper and lower
arm voltage and on the control of the
overall energy stored in converter leg.

Copies: [2]
Pages, total: [83]
Appendix: []
Supplements: []

By signing this document, each member of the group confirms that all group members
have participated in the project work, and thereby all members are collectively liable for
the contents of the report. Furthermore, all group members confirm that the report does
not include plagiarism.

3
4
Preface
This work, carried out by Giacomo Casadei, is a part of the final thesis to accomplish
the Master Degree in Automation Engineering at the University of Bologna (Italy). It
started at the beginning of February 2012 at the Department of Energy Technology,
Aalborg University, and will be completed in Winter 2012 at the University of
Bologna, Department of Electrical Engineering.

Prof. Remus Teodorescu, and Prof. Luca Zarri are the supervisors of this thesis.

Aalborg, June 8, 2012

5
6
Acknowledgements
I would like to thank my supervisor Prof. Remus Teodorescu for giving me the
opportunity to be involved with such a subject that is interesting and modern. I also
would like to thank Prof. Luca Zarri for its valuable help and guidance during my
study in Automation Engineering. Their advices and comments not only helped me
for specific issues but also improved my knowledge.

Additionally the author would like to thank all colleagues met in AAU, with whom he
shared this wonderful experience and work, and all his “home” and Danish friends.

Last but not the least, the author would like to thank all his family for infinite support
and love.”I’ll never be able to give you back what you gave me”.

A special thanks goes to Earth and Pillars.

Dedicated to silence.

Aalborg, June 8, 2012

7
8
Summary
The aim of this project is the analysis of a Modular Multilevel Converter (MMC) and
the development of a control scheme for energy stored. The converter is characterized
by a modular arm structure, formed by a cascade connection of a large number of
simple chopper cells with floating DC capacitors: these cells are called Sub-Modules
(SM) and can be easily assembled into a converter for high- or medium-voltage power
conversion systems. The analysis is based on the use of a simplified circuit constituted
by a single leg of the converter, where all the modules in each arm are represented by
a single variable voltage source. The circuit model is derived as a system of
differential equations that can be used for analyzing both the steady state and dynamic
behavior of the MMC, from voltages and thus energy point of view. The converter
structure requires an arm voltage balancing control and a leg total voltage control in
order to achieve a stable behavior in all operating conditions. The validity and the
effectiveness of the voltage control strategy are confirmed by numerical simulations.
Thesis objective is to deepen control issues about MMC family of converters: in order
to approach properly control system, a state of the art analysis had to be done.
This analysis showed that, as of today, few modeling approaches have been
investigated and thus, control issues are still difficult to be solved. In fact, it became
clear that a different modeling approach will be necessary in order to solve all
dynamic operation requirements.
Following a reliable modeling approach, carried out by A. Antonopoulos, L.Angquist,
and H.P. Nee in “On dynamics and voltage control of the modular multilevel
converter” (European Power Electronics Conference (EPE), Barcelona, Spain,
September 8-10, 2009), dynamic behavior of converter voltages and energy has been
analyzed.
New and improved regulators have been implemented: in fact, in order to solve
dynamic transients of the MMC (for instance LVRT, Failure Mode condition etc.) it
became clear that conventional control structures are not sufficient.
The over mentioned modeling approach leads to a highly non-linear system: non-
linearity is present also in the input, thus advanced control approach are difficult to be
implemented.
Thus, in the fifth chapter of this thesis, a new modeling approach is proposed, aiming
to simplify advanced control implementation.
It can be concluded, that with the new modeling approach a significant contribution to
control issues will be possible.

9
10
Thesis Layout
Chapter I

1. Introduction and Motivations

1.1. Objectives
1.2. Limitations

Chapter II

1. Description and Operation Principle of MMC

Chapter III

1. Mathematical Model of the MMC

2. Control Strategy

3. Consideration About Control Strategy And Model

Chapter IV

1. Development of a MATLAB model

2. First Model: Converter Operation and Model Verification

3. Second Model: Energy Control:

3.1. Simulation Results

4. Third Model: Output Power Control:

4.1. Simulation Results

5. Fourth Model: Mega-Watt System

5.1. Capacitance Relationship with Voltage Ripple
5.2. Simulation Results
5.3. Simulation Results: Sub-Modules Failure Mode
5.4. Simulation Results: LVRT

11
Chapter V

1. Motivations for a New Modeling Approach

2. Modeling Principles and Equations
3. Control Proposal
4. Future Works

Chapter VI - Conclusions

12
List of symbols – First Model
: sub-modules capacitor

: number of sub-modules in each arm

: total capacitance of the arm

: equivalent capacitance of the arm m

: arm impedance

: general modulation index

: m arm voltage

: voltage stored in arm capacitor

: current flowing through the arm

: upper and lower arm currents respectively

: output current

: differential current

: upper and lower sub-modules voltages respectively

: upper and lower modulation indexes respectively

: DC voltage

: output voltage

̂ : modulation carrier, with ̂ as pick value

: upper and lower arm voltages respectively

: differential voltage

: internal alternating voltage

: total energy stored in upper and lower sub-modules respectively

: total energy stored in one phase and difference between upper and lower

energy

13
: grid voltage

: line impedance

14
List of symbols – Second Model
: upper and lower arm voltages respectively

: output voltage

: upper arm impedance

: lower arm impedance

: upper and lower arm currents respectively

: output current

: DC voltage

: upper and lower modulation indexes respectively

: upper and lower arm capacitor voltages

: upper and lower arm equivalent capacitor

: upper and lower arm reference currents

: output current reference

: differential current reference

: upper and lower modulation indexes reference respectively

: upper and lower arm voltages reference respectively

: total energy stored in the arm and difference between upper and lower arm

energy

: total energy and difference of upper and lower arm energy references

15
16
Chapter I

1 – Introduction and Motivations

The development of new technologies and devices during the 20th century enhanced
the interest in electric power systems. Modern civilization based his operation on an
increasing energy demand and on the substitutions of human activities with complex
and sophisticated machines; thus, studies on electric power generation and conversion
devices become every day more and more important.
The recent attention in environment protection and preservation increased the interest
in electrical power generation from renewable sources: wind power systems and solar
systems are diffusing and are supposed to occupy an increasingly important role in
world-wide energy production in coming years.
Not only house utilities, but industrial applications and even the electrical network
requirements display the importance that energy supply and control will have in the
future researches.
As a consequence, power conversion and secondly control is required to be reliable,
safe and available in order to accomplish all requirements, both from users and legal
regulations, and to reduce the environmental impact.
Voltage Source Converter (VSC) technology is becoming common in high-voltage
direct current (HVDC) transmission systems (especially transmission of offshore
wind power, among others). HVDC transmission technology is an important and
efficient possibility to transmit high powers over long distances.
The vast majority of electric power transmissions were three-phase and this was the
common technology widespread. Main advantages for choosing HVDC instead of AC
to transmit power can be numerous but still in discussion, and each individual
situation must be considered apart. Each project will display its own pro and con
about HVDC transmission, but commonly these advantages can be summarized:
lower losses, long distance water crossing, controllability, limitation short circuit
currents, environmental reason and lower cost.
One of the most important advantages of HVDC on AC systems is related with the
possibility to accurately control the active power transmitted, in contrast AC lines
power flow can’t be controlled in the same direct way.
However conventional converters display problems into accomplishing requirements
and operation of HVDC transmission. Compared to conventional VSC technology,
Modular Multilevel topology instead offers advantages such as higher voltage levels,
modular construction, longer maintenance intervals and improved reliability.

17
 Figure 1 - ABB DolWin Installation, using HVDC connection

A multilevel approach guarantees a reduction of output harmonics due to sinusoidal

output voltages: thus grid filters become negligible, leading to system cost and
complexity reduction.
Like in many other engineering fields, modular and distributed systems are becoming
the suggested topology to achieve modern projects requirements: this configuration
ensure a more reliable operation, facilitates diagnosis, maintenance and
reconfigurations of control system. Especially in fail safe situations, modular
configuration allows control system to isolate the problem, drive the process in safe
state easily, and in many cases allows one to reach an almost normal operation even if
in faulty conditions.
However, it is necessary to take into account also disadvantages brought by Modular
design: the control scheme actually results more complex and requires a “two level”
configuration. Each modular part as to be controlled independently, but a master
control layer is needed to reach the overall system behaviour requirements. It is
necessary to provide a proper division of tasks between local controllers and master
controller: must of all, the communication system becomes really important in order
to achieve overall stability and time deadline accomplishment.
In the case of MMC, the concept of a modular converter topology has the intrinsic
capability to improve the reliability, as a fault module can be bypassed allowing the
operation of the whole circuit without affecting significantly the performance.
Many multi-level converter topologies have been investigated in these last years [1],
having advantages and disadvantages during operation or when assembling the
converters. To solve the problems of conventional multi-level converter a new MMC

18
topology was proposed in [2]-[5], describing the operation principle and performance
under different operating conditions. A simple schematic of this converter with N
modules per arm is shown in Fig. 1. The MMC proposed in [2]-[5] is one of the most
promising power converter topology for high power applications in the near future,
particularly in HVDC links (e.g. transmission of offshore wind power, among others).
Siemens has a plan of putting this converter into practical applications with the trade
name “HVDC-plus”. The system configuration of the HVDC-plus has a power of 400
MVA, a dc link voltage of 200 kV, and each arm composed of 200 cascaded chopper
cells [6].
This converter topology has been investigated by several research teams lately [7]-
[12]. These papers where mainly focused on the analysis of simplified circuit models
and improved control systems to achieve a reliable and stable operation of the
converter, as the cascaded connection of multiple sub-modules in each arm and leg
requires the voltage balance among the several sub-modules of each arm and between
upper and lower arms. The control of the total leg voltage and differential voltage
between upper and lower arms is a crucial point as it can affect seriously the operation
of MMC if not properly implemented. The potential interaction between these two
loops of control will be deepened and discussed; it will be important to evaluate the
possibility of control task distribution; this analysis is required to design the most
suitable control scheme.
The aim of this paper is to accomplish the stable voltage control of the MMC in all
operating conditions and the theoretical analysis is based on the circuit model
proposed in [8], hence, the same terminology will be used. The approach is based on
using a continuous model, where all modules in each arm are represented by variable
voltage sources, and PWM effects are neglected. The numerical simulations of the
converter show the presence of high currents that can circulate through the phase legs,
leading to the need of over-rating the modules. Besides to this, the presence of these
currents produces an energy transfer between the arms, leading to possible
instabilities of the converter. A suitable control strategy has been implemented for
avoiding instabilities in all operating conditions. The validity and the effectiveness of
the voltage control strategy is confirmed by numerical simulations.

1.1 – Objectives
In this thesis, analysis, modeling and control of a 13 MW/20 KV MMC converter are
investigated. Thus, objectives can be summarized:

 MMC operation principles: state of art and actual achievement

 MMC modeling: analysis of modeling approaches available
 Modeling Development and Simulation Verification of a KW system

19
  Modeling Development and Simulation Verification of a MW system
 Control Strategies Analysis and Implementation: firstly on the KW system,
then applied and verified on the MW system
 New proposal for modeling and control

1.2 – Limitations
This thesis main objective is to find a performing control system for the energy stored
in the converter: even if MMC is a well-known topology of converter, few modeling
and control approach are available.
To carry out this primary analysis, a modeling approach has been chosen and
followed, and can be found in [8]: this approach showed to be the most promising,
both for the analysis of MMC operation and for the development of control structure.
However, in Chapter V of this thesis a new modeling approach is suggested, in order
to simplify control strategy study.
Being the energy monitoring and control the aim of this project, modeling neglects
low level operation and dynamics of the system: sub-modules are simplified with an
equivalent variable voltage source, ideally controlled.
This assumption, gives the opportunity to study the system from a proper point of
view in terms of energy: on the other side, sub-modules operation is hidden and all
issues connected are not dealt in this thesis.
A different approach should be investigated, in order to account also sub-modules
operation in the modeling and through this, in control strategy development.

20
Chapter II
1- Description and principle of operation of MMC

The typical structure of a MMC is shown in Fig. 2, and the configuration of a Sub-
Module (SM) is given in Fig. 3. Each SM is a simple chopper cell composed of two
IGBT switches (T1 and T2), two anti-parallel diodes (D1 and D2) and a capacitor C.
Each phase leg of the converter has two arms, each one constituted by a number N of
SMs. In each arm there is also a small inductor to compensate for the voltage
difference between upper and lower arms produced when a SM is switched in or out.

Figure 2 - Schematic of a three-phase Modular Multi-level Converter

With reference to the SM shown in Fig. 2, the output voltage UO is given by,

UO = UC if T1 is ON and T2 is OFF
UO = 0 if T1 is OFF and T2 is ON

where UC is the instantaneous capacitor voltage.

The configuration with T1 and T2 both ON should not be considered because it
determines a short circuit across the capacitor. Also the configuration with T1 and T2
both OFF is not useful as it produces different output voltages depending on the
current direction. Fig. 4 shows the current flows in both useful states.

21
In a MMC the number of steps of the output voltage is related to the number of series
connected SMs. In order to show how the voltage levels are generated, in the
following, reference is made to the simple three level MMC configuration shown in
Fig. 5.

T1 D1
UC

C
UO T2 D2

Figure 3 – Chopper cell of a Sub-Module

T1 OFF - T2 ON T1 ON - T2 OFF
SM off-state SM on-state

Direct current flow

Inverse current flow

Figure 4 - States of SM and current paths

22
 UD /2
VAC

UD /2

Figure 5 - Schematic of one phase of Three-Level Converter

In this case, in order to get the positive output, +UD/2, the two upper SMs 1 and 2 are
bypassed. Accordingly, for the negative output, - UD/2, the two lower SMs 3 and 4 are
bypassed. The zero state can be obtained through two possible switch configurations.
The first one is when the two SMs in the middle of a leg (2 and 3) are bypassed, and
the second one is when the end SMs of a leg (1 and 4) are bypassed. It has to be noted
that the current flows through the SMS that are not by passed determining the
charging or discharging of the capacitors depending on the current direction.
Therefore, in order to keep the capacitor voltages balanced, both zero states must be
used alternatively. The voltage waveform generated by the three level converters is
shown in Fig. 6.

+UD /2

-UD /2

Figure 6 - Voltage waveform of a Three-Level Converter

23
The principle of operation can be extended to any multi-level configuration as the one
represented in Fig. 7.

UD /2

VAC

UD /2

Figure 7 - Schematic of one phase of Multi-Level Converter

In this type of inverter, the only states that have no redundant configurations are the
two states that generate the maximum positive and negative voltages, + UD/2 and
–UD/2. For generating the other levels, in general there are several possible switching
configurations that can be selected in order to keep the capacitor voltages balanced. In
MMC of Fig. 6, the switching sequence is controlled so that at each instant only N
SMs (i.e. half of the 2N SMs of a phase leg) are in the on-state. As an example, if at a
given instant in the upper arm SMs from 2 to N are in the on-state, in the lower arm
only one SM will be in on-state. It is clear that there are several possible switching
configurations. Equal voltage sharing among the capacitor of each arm can be
achieved by a selection algorithm of inserted or bypassed SMs during each sampling
period of the control system. A typical voltage waveform of a multi-level converter is
shown in Fig. 8.

+UD /2

-UD /2

Figure 8 - Voltage waveform of a Multi-Level Converter

24
Chapter III
3.1 - Mathematical model of the MMC

The typical structure of an MMC, shown in Fig. 1, can be summarized into 3 levels:

I. sub-modules SM (the lower level, usually Chopper cells)

II. arm (second level of the converter, half of the leg-phase)
III. leg (can be considered one phase)

If a two-level control structure is considered, for instance a Master-Slave structure,

lower level of control is assigned to sub-models (level I), instead upper level of
control deals with arm-leg voltage and currents control (level II, III). It is assumed
that a lower level control for the sub-modules is present ensuring that all capacitors
are equally charged.
As suggested in [13], most of existing investigations and simulations are based on
switched or discrete models. However, discrete models have two main disadvantages:

a) discrete models do not allow an analytical approach to model the converter

and to design the control system;
b) numerical solution of complex converter configurations using a high number
of SMs require considerable simulation time.

A continuous model can overcome these disadvantages. Therefore, to clearly

understand the operation of this converter it is necessary to write the voltage-current
equations and to determine a continuous model suitable to design a control scheme.
In the following, an analysis is carried out with the aim to control the upper and lower
arm voltages, using the continuous model presented in [8], that considers only one
converter phase and is based on time-variable capacitors.
Considering a converter with N sub-modules per arm, each arm can be controlled with
an insertion index (modulation index) n(t), where n(t)=0 means that all sub-modules
in the arm are by-passed, on the other side n(t)=1 means that all the sub-modules in
the arm are inserted.
Ideal capacitance of the arm should be:

25
The effective capacitance of the arm is dependent on the insertion index, so it can be
written as:

where the m apex means the number of the arm (for instance, in a three-phase
converter m=1,2,3,..,6).
It would be possible to have a full representation of the MMC converter, including
operation of each sub-module, but this approach tends to be fairly complicated and
not easy to be used as a base for control schemes development.
A simpler way would be to consider a continuous model, but 2 important assumptions
are necessary in order to develop this approach:

1. the switching frequency is much higher than the frequency of the output
voltage
2. the resolution of the output voltage is small, compared to the amplitude of the
output voltage (i.e. high number of sub-modules)

Assuming 1 and 2, it’s possible to create a continuous model which represents the
overall operation of the converter, neglecting the single sub-modules behaviour: this
type of model is suitable for control system design and makes it possible to focus on
the energy stored in the converter and its balance between arms. The simplified model
is shown in Fig. 9.

Carm
UD/2
ucu
iU

Zarm
iV

Zarm

iL
Carm
UD/2 ucu

Figure 9 - Simplified Circuit Used for the Analysis

26
Again, apex m represent the arm, n(t) is the insertion index, instead ucΣ(t) is the sum
of all capacitance voltage in the m arm. Then equations (3) and (4) follow.

With reference to Fig. 8, where only one phase is considered, it is possible to write a
set of equations for currents: currents from upper and lower arms, rispectively iU and
iL, will constitute the output current iV.
The idiff current represents the current that circulates from the phase leg to the DC link
(and/or to another phase leg).

These equations are representing the ideal condition in wich the contributions of
upper and lower arms to the output current are equal. The difference current is
introduced to consider the possible situation in which the capacitors are not equally
charged to the reference value. In this MMC configuration the sum of all capacitor
voltages of one arm is assumed to be equal to the DC Voltage uD.
Equations (6) and (7) are the same of (4), just emphasizing the contributions of upper
and lower arms in terms of voltage and current.

With all these equations and the semplified circuit shown in Fig. 8, it is possible to
write equations (8) and (9) as simple Kirkhoff voltage equations.

27
Subtracting equation (9) to (8), and substituting the following equations

it is possible to obtain this dynamic equation of the current idiff :

From equations (6) and (7), substituting iU and iL with the expressions given in (5),
two dynamic equations of upper and lower arm voltages are obtained:

From (12) it can be noted that with idiff equal to zero, the load current acts in order to
unbalance the upper and lower arm voltages. In steady state conditions the load
current is changing assuming positive and negative values, then, the time derivative of
the arm voltages are also changing, and in ideal conditions the arm voltage should
oscillate around a constant mean value. The presence of non idealities and losses may
lead the converter to be unstable. As a consequence, it can be concluded that only the
presence of a suitable difference current allows the converter to operate correctly.
Using (11) and (12), a dynamic and continuous model is thus obtained and shown in
(13).

[ ] [ ]

[ ] [ ]

Assuming that a sinusoidal output voltage is desired, the reference signal for
modulation is

28
 ̂

and the ideal terminal voltage is given by

In order to solve the system of equations (13) the load current should be known. Here,
the following alternating current is assumed as the output load current

Where φ is the load phase angle and is arbitrary.

Using the reference signal in (14), the modulation indices for upper and lower arms
can be expressed as

It can noted that the sum of upper and lower modulation indexes is always equal to 1.
As a consequence, assuming the capacitor voltages of all modules equal to the
refernce value, the sum of the voltages of upper and lower arms is always equal to the
DC voltage uD.
In real operating conditions the capacitor voltages will not be exactly equal to the
reference value and as a result a difference voltage will be present forcing a difference
current to flow between the leg and the DC source. A suitable control of this current is
crucial for achieving a correct operation of the converter and an equal sharing of the
DC voltage among all modules.
A possible control strategy is the one based on adding an offset voltage to upper and
lower arm voltages uCU and uCL defined trough (17) and (18). This offset voltage (udiff)
is determined with some criteria aimed to keep the module voltages as close as
possible to the reference value or to keep the energy stored in upper and lower
capacitors equalized.
It will be shown in the next section that udiff will not affect the terminal voltage (as uV
is related to the difeerence between upper and lower arm voltages), but will impact on
idiff current instead.

29
The number of levels that can be obtained depends on the assumptions made for the
analysis. When assuming a constant DC voltage, actually it is possible to generate
output voltages having a number of levels equal to 2N+1, whereas the number of
levels must be reduced to N+1 if the the DC voltage has to be kept under control by
the converter itself.
This is the situation that occurs in HVDC systems composed by two MMCs
connecting the two ends of a DC cable. In this case there is no DC capacitor between
the DC voltage terminals and a DC voltage controller is necessary. In this analysis a
constant DC voltage will be assumed as the aim is to develop a strategy for keeping
under control the total energy stored in each leg and the unbalance between the energy
stored in upper and lower arms.

3.2 - Control Strategy

By adding and subtracting equations (9) to (8), it is possible to obtain two equations
that clearly explain a possible approach to MMC control:

As suggested in [8], from (19) and (20) is possible to draw the following conlcusion:

 the output voltage uv depends only on the current iV and on the difference
between upper and lower voltage uCU, uCL
 the arm voltage difference acts like an inner alternating voltage, R and L as a
passive inner impedence for alternating current
 idiff depends only on the DC link voltage and the sum of arm voltages

Therefore it is possible to note that adding the same quantity to both arm voltages will
not affect the AC side, but will influence the difference current instead, wich can thus
be controlled.
It is opportune for control purposes to obtain an expression of upper and lower arm
voltages in wich this voltage difference is included. Starting from (8) and substituting
the upper current with

30
leads to

(21) can be rewritten as

Considering that form (19)

(22) can be rewritten as

The same development can be carried out for uCL yielding

where udiff is defined as

Equations (24) and (25) show how the difference voltage contribute to the reference
values of upper and lower arm voltages.
The quantity ev defined as

31
is practically representing the output voltage uv, apart from the voltage drop across R
and L as shown in (19). So, in the following, ev is used to represent the output voltage
uv.
In order to determine a possible control strategy for determining udiff , it is opportune
to introduce the energy stored in upper and lower arm capacitors.
Assuming that voltage and thus energy stored in each arm is equally shared between
sub-modules, the upper and lower energy are

[ ( ) ]

[ ( ) ]

Considering now power equations of both upper and lower arms, it is possible to
write:

( )( )

( )( )

The total capacitor energy in the leg and the difference capacitor energy between
upper and lower arms are

Differentiating (32) and (33), and introducing (30) and (31) yields

( )

( )

32
These equations are very important to discuss the influence of idiff on the total and
difference capacitor energies. From (34), assuming idiff having only a dc component its
product with uD represents the power delivered to the ac side (load and capacitors).
The quantity udiff idiff represents the losses on the arm resistance R and the magnetic
energy variation in the arm inductance L, is the load power. So, a DC component
of idiff can be used to control the total capacitor energy.
From (35), it can be noted that a DC component of idiff has no impact on the difference
capacitor energy as there are no dc components in ev. The conclusion is that the DC
component of idiff can only be used for controlling the total capacitor energy stored in
the converter leg.
On the other hand, an alternating component of idiff, having the same fundamental
frequency as the output voltage ev, could be usefully employed to control the
distribution of the capacitor energy between upper and lower arms. In fact, the
product ev idiff has a dc component that can be used to force the energy difference to
change. A similar effect is created by udiff iV, but this quantity should be smaller for
small value of R and L, thus in the following the control strategy will be developed
with reference to the quantity ev idiff only.

3.3 - Consideration about Control Strategy and

Model

It is necessary to make an important remark: the equations wrote and discussed in

previous section lead to a continuous model, suitable for analysis and understanding
of operation principle of the MMC. On the other hand, from the control point of view,
these equations are not easy to use: even if it is possible to decouple continuous and
alternate component of differential current, in order to properly track references, non-
linear couplings make really complex advanced control structures.
From equations (34) and (35) it becomes clear that the input variables, or rather
modulation indexes, influence dynamics of the energy as a non-linear input: in fact,
the modulation indexes directly influence differential voltage, and the relationship
between differential voltage and differential current is a low-pass filter (Fig. 10).

Figure 10 - Input Control Relationship

33
Neglecting this low pass filter it is possible to consider that input variables are a
“square” input for the system: this condition makes more difficult the development of
advanced control strategies and the study of complex control strategies.
Thus, a different approach in the modeling will be developed in order to simplify the
analysis from the control point of view: this modeling approach will be deepened in
Chapter V.
The control strategy implemented is a linear control which aims to operate in the
proximity of the region where the control can be considered linear: actually, control
input will be limited in order to preserve this assumption. Otherwise, the non-linear
input of the control can affect operation of the system, leading to instability.
Two different loops will be implemented, the first one to control the overall energy of
the MMC leg, the second to control the balance between upper and lower arms of the
phase-leg.
The interaction between two loops may lead to instability: because of the non-linear
configuration of system equations, it is hard to analyze systems coupling properly (for
instance Relative Gain Array analysis).
It is however evident that total energy and energy balance interact dynamically in
system operation: in order to make a decoupling, balance of energy loop is tuned in
order to be greatly slower than the overall energy loop. This frequency decoupling
will highly benefit differential current waveform: if not performed, overall energy and
balance interaction leads to a really distorted differential current. This current,
flowing in the phase-leg, would produce a voltage drop on the phase impedance,
increasing losses and disturbances in the stability of the system.
Then, an output current control loop will be implemented in order to simulate the
operation of the converter in all different possible conditions. The general block
diagram of control scheme is shown in Fig. 11.

34
Figure 11 - Control Scheme, Block Diagram

35
36
Chapter IV

4.1 - Development of a MATLAB model

Starting from equations and discussions of Chapter III, it is possible to define a simple
dynamic model of one phase of the converter.
Below are listed the chosen equations:

[ ] [ ]

[ ] [ ]

( )( )

( )( )

Equation (36) is the output voltage equation, where ug is the grid voltage, and Rg, Lg
the parameters of the line connecting MMC to the grid.
The previous equations are implemented on Simulink, with the use of Embedded
MATLAB Functions.

37
4.2 - First model: Converter Operation and Model
Verification

The first system implemented in MATLAB-Simulink includes only the equations in

(13), but it is however important to understand the operation of the converter: in
particular, the necessity of controlling the differential current and thus the energy
stored in upper and lower arm becomes clear.
In this simulation, output voltage and current are imposed and the operation of the
system is then observed: modulation indexes are calculated ideally, using equations
(24) and (25), neglecting differential voltage usage (no control strategy is
implemented yet) and neglecting also the voltage drop of the output current on the
arm parameters (i.e. ).

i_dif f

[nu] nu

From7 Scope

[nl] 1 [i_diff]
nl i_dif f _der
s
fcn
From10 Integrator i_diff
[u_v] u_dif f
[u_cu] u_v
u_cu

From8 From18

[u_cl] u_cl
Scope4

[u_cu] nu [nu]
From11 u_cu
I_DIFF
From12 u_cu1

[i_diff] i_dif f
[u_cl] nl [nl]
u_cl
From 1 [u_cu] u_cl1
[nu] nu s From13
Integrator1 u_cu Control System
From2 U
fcn
[nl] nl 1
[u_cl] Scope2
s
From3
[i_v] Integrator2 u_cl
i_v

From9 U_CU & U_CL

Scope1

[nu] nu [i_v] i_v [i_u]

From4 From1
[u_arm_up] I Goto4
[nl] nl fcn
Goto2 [i_diff] i_dif f [i_l]
From5 U_arm
fcn From6
[u_cu] u_cu [u_arm_low] Goto5
I_u & I_l
From14
Goto3
[u_cl] u_cl

From15
U_arm Calculation
with U_CL,U_CU and
Modulation Indexes
Scope3
[i_v]

Output Current (pi/36 phase) Goto

Ud/4 [u_v]

m(t) Gain6 Goto1

Scope5

Figure 12 - First Model MATLAB Scheme

MATLAB Model Implementation can be seen in Fig. 12.

Green blocks implement model equations; the red block instead contains the
calculation of the modulation indexes: in the right bottom, output voltage and current
are imposed, assuming a load-phase angle equal to zero.

38
The upper green block represents the dynamic of differential current; the middle green
block represents the capacitor voltage dynamic, the lower block calculates the upper
and lower arm voltage, multiplying the capacitor voltage by the modulation index. On
the right side, the green block is used to calculate the upper and lower arm current.
Output voltage and current, and other system parameters are listed below in Tab. 1.

Tab. 1 – List of system parameters

It is important to underline that the model implemented is a kW system: the choice to

preliminary study a smaller system, compared with the final objectives of project, is
motivated by control issues and parameters availability.
It will be shown that control strategy implemented is almost independent from
parameters changes; however it is easier to start solving the problem from a small
scale system. Furthermore, parameters for a kW system were available from the
beginning of the project, instead MW parameters required some time and deepening
to be confirmed.
No control loop is implemented yet, and the converter is not able to keep the capacitor
voltage to the initial value, i.e. the capacitor voltage decreases from the initial value to
a non-desired “steady state condition”, the same happens to differential current.
Arm voltages waveforms are shown in Fig. 13.
After a brief transient the differential current acts in order to maintain the capacitor
voltage stable (waveform in Fig. 14), and the system is operating properly.
In order to control the capacitor voltage and thus totla energy stored in upper and
lower arm, and their balance, a control on the differential current must be
implemented.

39
 210

200

190

180

170

160

150

140

130
0 1 2 3 4 5 6 7 8 9 10
Time (s)
Time
Figure 13 – Upper (Blue) and Lower (Green) Arm Capacitance Voltage (V):
without control loop

2.5

1.5

0.5

0
0 1 2 3 4 5 6 7 8 9 10
Time
Time (s)
Figure 14 - Differential Current Waveform (A): without control loop

40
4.3 - Second model: Energy Control
The second system implemented includes energy equations (30) and (31) and propose
an energy control strategy, to obtain a stable single-phase leg of the converter: to
accomplish this, differential current is used. Like the previous model, output voltage
and current are still imposed and the operation of the converter-leg is consequently
observed and controlled.
As deducted by equations (19) and (20), differential current can be controlled by a
differential; this voltage will not affect the AC (output) side (i.e. adding the same
quantity to upper and lower arm voltage does not change the output current or
voltage, but imposes a differential current to flow in the phase-leg).
Thus, a simple but efficient control strategy controls overall energy stored and the
balance between upper and lower arm energy through the differential current, using
the differential voltage as intermediate control input: in fact this differential voltage,
sum of two different control components, will modify modulation indexes, changing
the ideal value calculated assuming constant the capacitor voltages.

Controlling The Overall Energy Stored In the Converter

i_dif f
and The Balance of Energy

[nu] nu

From7

[nl] 1 i_diff
nl i_dif f _der
s
fcn nu [nu]
From10 Integrator i_diff
u_cu u_cu1
u_cu

From8

u_cl u_cl
Scope4

From11
I_DIFF Scope
Scope6

i_diff i_dif f

From 1 nl [nl]
u_cu
[nu] nu s
Integrator1 u_cu u_cl1
From2 U_der
fcn
[nl] nl 1 u_cl Scope2
s
From3
i_v Integrator2 u_cl
i_v Control System
From9 U_CU & U_CL
Scope1
Scope7

[nu] nu i_v i_v [i_u]

From4 From1
[u_arm_up] I Goto4
[nl] nl fcn
Goto2 i_diff i_dif f [i_l]
From5 U_arm
fcn
u_cu u_cu From6 Goto5
[u_arm_low]
I_u & I_l
From14
Goto3
u_cl u_cl

From15
U_arm Calculation
with U_CL,U_CU and
[u_arm_up] Scope8
Modulation Indexes u_arm_up
Scope3
i_v From21
w_cu
Output Current (pi/36 phase) Goto [u_arm_low] u_arm_low
1 W-cu
From20 W
s
fcn
Ud/4 u_v i_diff i_dif f Integrator3 w_cl
m(t) Gain6 Goto1 From17
w_cl
i_v i_v Scope11

From19
W_CU & W_CL

Scope5 Scope9

Scope12

Scope10

Figure 15 - Second Model MATLAB Scheme

41
Dynamic equations for upper and lower energy are implemented in the lower, right
side MATLAB Embedded block: then, the sum of the two energies stored in upper
and lower arms is the overall energy and a constant reference value is imposed. The
difference of upper and lower arm energies is used to control the balance between
upper and lower arm; thus, the reference value is set to zero. The control scheme
block is then explained and can be seen in Fig. 16.
Red blocks implement the overall energy loop: a PD controller works on the total
energy error, with the purpose of deleting the error between energy reference and the
energy effectively stored in the phase-leg.
However, being all the imposed output signals sinusoidal, there will be a steady state
error in the overall energy: the goal is to reduce this error as much as possible and to
have a fast dynamic response in case of change reference.
Green blocks implement the balance control between upper and lower arm: it is again
a PD controller, but this time a low pass filter is also required, in order to have a
stable and smooth loop. The control action generated by this loop is then multiplied
by a “carrier”, with the same frequency of the output signals and the impedance arm
phase.
From equations (34) and (35), rewritten in the following for convenience, it is
possible to understand how to control properly the overall energy and the energy
balance.

( )

( )

In equation (34), differential current multiplies the DC voltage, so a DC component of

differential current is used to control the overall energy stored.
In equation (35), differential current multiplies the output voltage ( can be
considered the same of output voltage, neglecting the voltage drop of output current
on arm parameters), so an AC component of differential current is used in order to
control the balance between upper and lower energy (using the carrier technique
implemented by the yellow block).
The two control components are added and then used as a differential voltage
reference which will be followed by changing upper and lower modulation indexes,
driving the necessary differential current to control the system energy: modulation
indexes are calculated dividing the upper and lower voltage (calculated with equations
(24) and (25)) by the capacitor voltage reference, as it is shown in the lower part of
the control scheme with green blocks.

42
Figure 16 – Control Scheme Implementation in MATLAB

43
An important consideration must be performed: both loops implemented to control
energy behavior include saturation. There are two important reasons in order to
impose these limitations:

 Stability of the System: it is necessary to leave the control enough slack in

order to properly act on the system. It is however important to remember the
non-linear configuration of the input, which impose a strict constraint to input
magnitude
 Output Tracking: this point will be shown clearly in the following model,
Chapter 4.4, where output variables control is implemented. It is important to
use a small percentage of voltage available to control energy: the remaining
part is necessary to control the output.

4.3.1 - Simulation Results

Imposing the output current, the overall energy stored in one phase leg has a brief
transient after which it oscillates around the steady state reference value: the
amplitude of the oscillation is less than 1% and the frequency is the same of the
output signals. The waveform is shown below, in Fig. 17.

201.5

201

200.5

200

199.5

199
0 1 2 3 4 5 6
Time
Time (s)
Figure 17 - Overall Energy Stored in the Converter Phase (Joule): steady-state condition
of the overall energy in a converter phase

44
Initial transient is caused by the difficulty in finding a proper initial condition for the
system: even if the capacitors are charged, also differential current has to be set to a
proper value. Even if the mean value can be found, the alternate component of the
differential current in order to balance energy/voltages requires a transient.
The energy difference between upper and lower arm has a similar behavior: after a
brief transient, it reaches a steady state oscillation around the reference, 1% of
amplitude compared to the overall energy value and with the same frequency of the
output (Fig. 18).

-1

-2

-3
0 1 2 3 4 5 6
Time
Time (s)

Figure 18 - Difference Between Upper and Lower Arm Energy (Joule): steady-state
condition of the difference energy

The steady state behavior is stable as expected: some problems may occur, especially
if dynamic changes of the energy are imposed.
In these cases, transient with oscillations are present and they will be discussed and
deepened in coming sections.

45
4.4 - Third model: Output Power Control

With the third implemented model, the goal is to deepen the behavior of the converter
about the output current control: proved the overall and balance energy control
stability, it is now possible to introduce the control of the output current.
To track the output current, a resonant controller is used; a standard PI structure
would be insufficient to cancel the sinusoidal error, so a different approach is
necessary to be adopted. The resonant controller is the most suitable for a single-
phase system: however, considering the three-phase general structure, a D-Q or Space
Vector transform will be chosen for the control structure, in order to simplify the
complexity of control loops.
A MATLAB Embedded block is added in order to implement the output current
dynamic equation; then, a resonant controller is applied to output current error, as it is
shown in Fig. 19. The controller is composed of two main contributions: a
proportional, and a resonant part. The Bode diagram is shown below in Fig. 20. The
resonant peak is in correspondence of the frequency of the output and can be changed
dynamically in order to follow different outputs (in terms of frequency).

1 Kp _i_r 1
i_v_err e_v_ref
Kp

Ki _i_r.s
s2 +omega 2

TF 1

Figure 19 - Resonant Controller MATLAB Scheme

Bode Diagram

75

70

65
Magnitude (dB)

60

55

50

45

-45
Phase (deg)

-90

-135

-180
2.41 2.43 2.45 2.47 2.49 2.51 2.53 2.55 2.57 2.59
10 10 10 10 10 10 10 10 10 10
Frequency (rad/s)

Figure 20 - Bode Diagram of Resonant Controller

46
The output of the resonant controller is a reference for the inner alternate voltage ,
which again will be achieved through the control of modulation indexes.
The general scheme implemented in MATLAB for the third model is represented in
Fig. 21.

Figure 21 - Third Model MATLAB Control Scheme

47
4.4.1 - Simulation Results

Simulation results show that adding control loop for the output current does not
impact the energy behavior of the system.
Both overall energy and the balance between upper and lower arm behave as shown in
previous examples: thus, it is possible to consider that energy control and output
current control are decoupled. This is possible because of saturation imposition:
actually, limitations and constraints on differential voltage are important both for
stability of the system and output variables tracking.
If tuned properly, the energy control system will use a small fraction of the DC
voltage; the remaining part is used to guarantee the output tracking. Differential
current control exploits only a small quantity of the voltage available; the rest is used
to control the output.
If the trade-off between the energy control and output power control is properly tuned,
energy loop and output loop can be considered decoupled.
Fig. 22, shows the behavior of differential current; the current has a DC component of
1.1 A, and an alternating component around 0.1 A of amplitude.

1.5

1.4

1.3

1.2

1.1

0.9

0.8

0.7
0 1 2 3 4 5 6
Time (s)
Time
Figure 22- Differential Current Waveform (A)

48
The error of the output current is shown in Fig. 23; it is possible to see that, after a
brief transient, the current reaches almost the desired value with an error less than 1%
of the steady state value of the output current.
1

-1

-2

-3

-4

-5
0 1 2 3 4 5 6
Time
Time (s)

Figure 23 - Output Current Error (A)

0.1

-0.1

-0.2

-0.3

-0.4
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time
Time (s)
Figure 24 - Zoom of Output Current Error (A)

49
Fig. 24 shows a zoom of the fast transient of output current error, almost impossible
to be seen in Fig. 23. The initial error is of course big (the output current starts form a
null value), but after few milliseconds the error is less than 1% of the steady state
value.
The dynamic behavior of the system needs to be investigated. To this aim, a dynamic
change of the overall energy is imposed in order to show the behavior of the system.
The imposed change is not connected to any particular working situation; it is just
used to show the dynamic response in time and damping, obtained with the control
system. For instance, this increase of the overall energy may be related to a failure
mode condition, considering the increase of energy a way to maintain the system
stable even if a certain part of the converter is not working properly.
Simulation results, illustrated in Figs. 25 and 26, show the behavior in case of a
simple proportional regulator: actually, is not a usually accepted operation response in
dynamic systems. Even if the response itself is fast, the damping obtained with a
proportional controller is not satisfying, especially in the overall energy control loop.
These results call for a PD configuration of the controller: adding a derivative action,
enough phase margin can be reached, in order to smooth the oscillation of the system.

300

280

260

240

220

200

180
0 1 2 3 4 5 6
Time
Time (s)

Figure 25 - Dynamic Response of Overall Energy Stored in Phase Leg (Joule): transient
generated by a step-change in the overall energy reference

50
 6

-2

-4

-6
0 1 2 3 4 5 6
Time
Time (s)
Figure 26 - Dynamic Response of the Difference between Upper and Lower Arm Energy
(Joule): overall energy oscillations affect also the balance loop

Derivative action is underlined with red background in Fig. 27 and simulation results
are shown in Figs. 28 and Fig. 29. Energy behavior improvements are evident and the
control system now works as expected.

1 Kp_t 1
W_tot _err u_diff _tot
Kp Saturation 2

s+D_zero
s+D_polo
TF 1 Saturation 3

Figure 27 - Addition of Derivative Action in the Controller

51
260

250

240

230

220

210

200

190
0 1 2 3 4 5 6
Time (s)
Time
Figure 28 - Overall Energy Behavior with PID Regulator (Joule)

-2

-4

-6
0 1 2 3 4 5 6
Time (s)
Time
Figure 29 - Difference Energy with PID Regulator (Joule)

52
4.5 - Fourth model – Mega-Watt System

The model presented and studied in previous sections was a kW power system: in this
chapter the analysis and simulation results will be proven on a MW power system.
The MW power system requires only changing the limitations on the differential
voltage available: as much as the power increases, as much the differential voltage,
used to control the energy stored in the system, has to increase.
The control system scheme remains exactly the same, and will be proven to be
reliable and tightly dependent with system parameters: however, an important
consideration has to be deepened about system project.
Even if not compromising stability, plays a fundamental role in the operation of
the system: the sizing of this parameter has to take in consideration the voltage ripple
that it’s considered to be acceptable. When considering the previous simulations (the
kW power system) was big enough in order to minimize voltage ripples in the
upper and lower modules: considering a MW power system, it is important to discuss
the size of the capacitor in order to understand the relationship between this parameter
and control constraints.

4.5.1 - Capacitance - Voltage Ripple Relationship

Starting from voltage derivate equations in (13), here rewritten for convenience,

it is possible to find a relationship between voltage and arm capacitance. Considering

only the upper arm voltage (but same results would be obtained considering lower
arm voltage), it is possible to substitute currents with their behavior over time

For the differential current, it is possible to consider only the DC component: this
component is used to maintain the capacitor voltage to a constant value.
It is possible to integrate the voltage equation in the positive half period of the output
current, yielding

53
 ∫ [ ]

where modulation index con be considered as

with the hypothesis that

(41)

Substituting (40) and (41) in (39) we obtain

∫ [ ]

(42)

[ [ ] [ ] ]

[ ]

Finally a relationship between and voltage ripple is found as

[ ] [ ]

With this equation it is possible to understand the system operation and thus to size
properly the capacitance of the arm, once assigned output current and calculated the
DC component of differential current, in order to satisfy voltage ripple constraints.

54
4.5.2 - Simulation Results
Parameters of Mega-Watt power system are listed in Tab. 2.

Tab. 1 – List of system parameters

The first simulation shown in the report shows the steady state operating conditions of
the MW power converter: as expected, the general behavior of the converter is
unchanged in comparison with the kW power converter.
The only change can be seen in the order of magnitude of variables observed. Figs. 30
and 31 show total energy and balance energy waveforms.
6
x 10
2.015

2.01

2.005

1.995

1.99

1.985

1.98
0 1 2 3 4 5 6
Time
Time (s)
Figure 30 - Overall Energy (Joule) Steady State Waveform

55
 4
x 10
3

-1

-2

-3

-4

-5

-6
0 1 2 3 4 5 6
Time
Time (s)

Figure 31 - Difference Between Upper and Lower Arm Energy (Joule)

420

400

380

360

340

320

300

280

260
0 1 2 3 4 5 6
Time
Time(s)

Figure 32 - Differential Current (A) in Steady State Conditions

56
In Fig. 32 differential current behavior is shown: continuous component increased
proportionally to output current and line voltage, in order to balance output power
increase. The alternate component remains limited and guarantees stability in upper
and lower arm voltage. It is possible to notice the DC component, around 340 A, to
balance output power and the AC component to balance Upper and Lower voltages.
Fig. 33 instead shows upper and lower arm voltages: as deepened in previous section,
oscillations of these voltages are related to capacitors designed for the converter. In
order to reduce the ripple it would be necessary to increase the capacitor in each sub-
module: however, with the actual design and control tuning, voltage ripple is less than
1%.
4
x 10
2.04

2.03

2.02

2.01

1.99

1.98

1.97

1.96
0 1 2 3 4 5 6
Time
Time (s)

Figure 33 – Upper (Blue) and Lower (Green) Arm Voltages (V) in Steady State
Conditions

The second simulation aims to test dynamic response of the system, in terms of step
variation of the overall energy stored in the converter. As in previous models, this
kind of simulation is not representative of a particular operating condition: it is
however the best way to test control stability and performances of the total energy
loop.
Even with a 50% change of the overall energy reference, the loop response respects
all requirements: rising time is around 500 ms and there is no overshoot or relevant
oscillations around the new steady state condition, as represented in Fig. 34.

57
The influence of the overall energy loop on the balancing control loop is almost
negligible, due to the frequency decoupling performed with control loops, and can be
seen in Fig. 35.
From Fig. 36, it can be noted that differential current waveform respects all
expectations: there is a short transient between initial and final steady state conditions
in order to accomplish the overall energy reference change
The same dynamic response can be seen in upper and lower arm voltages as
represented in Fig. 37.

6
x 10
3.2

2.8

2.6

2.4

2.2

1.8
0 1 2 3 4 5 6
Time
Time (s)
Figure 34 - Overall Energy (Joule) Step Response with a 50% Change of the Overall
Energy Reference

58
 4
x 10
4

-1

-2

-3

-4

-5

-6
0 1 2 3 4 5 6
Time
Time (s)
Figure 35 - Difference Between Upper and Lower Arm Energy (Joule) with a 50%
Change of the Overall Energy Reference

550

500

450

400

350

300

250
0 1 2 3 4 5 6
Time
Time (s)
Figure 36 - Differential Current (A) Behavior with a 50% Change of the Overall Energy

59
 4
x 10
2.5

2.4

2.3

2.2

2.1

1.9
0 1 2 3 4 5 6
Time
Time (s)
Figure 37 - Upper (Blue) and Lower (Green) Arm Voltages (V) with a 50% Change of
the Overall Energy

4.5.3 - Simulation Results: Sub-Modules Failure Mode

In order to properly test also the balancing loop and to verify the coupling with
overall energy loop further simulations are necessary: the initial condition of
capacitors is set so that the system starts from an unbalanced condition. Thus, it will
be possible to understand if the decoupling is possible in both directions of
interaction: previous simulations showed that overall energy loop slightly influences
balancing loop, but it is not possible to conclude the same for the opposite coupling.
This initial unbalance condition is related to a fault condition, where a certain number
of broken sub-modules are bypassed: in order to remain in a fail-safe region, the
control has to be fast and reliable, reaching the steady state condition without
overshoots or oscillations.
Fig. 38 shows the overall energy behavior: starting from an unbalanced initial
condition, the interaction between balancing loop and overall energy loop is not
negligible.

60
 6
x 10
4

3.5

2.5

1.5

1
0 1 2 3 4 5 6
Time
Time (s)
Figure 38 - Overall Energy (Joule): Response to Unbalanced Initial Conditions

The balancing loop acts slowly and takes a longer transient to set to zero the error, as
can be noted in Fig. 39. The overall energy loop instead acts immediately to reach the
steady-state reference value. As a consequence, the unbalanced initial condition leads
to dangerous operation due to an important overshoot of the overall energy stored in
capacitors, as clearly emphasized in Fig. 38.
The differential current flowing through the leg is the same for both upper and lower
sub-modules: even if only one of the arms needs to be charged, the differential current
will charge both of them, leading to a condition where the overall energy of the
system overshoots greatly the reference value. This can be clearly understood
analyzing the upper and lower arm voltage waveforms, shown in Fig. 40. It is
possible to notice the overshoot of booth voltages, due to the same charging current.

61
 5
x 10
1

-1

-2

-3

-4

-5

-6

-7

-8
0 1 2 3 4 5 6
Time
Time (s)
Figure 39 - Difference between Upper and Lower Arm Energy (Joule): System Response
to Unbalanced Initial Conditions

4
x 10
3

2.8

2.6

2.4

2.2

1.8

1.6

1.4

1.2

1
0 1 2 3 4 5 6
Time(s)
Time
Figure 40 - Upper (Blue) and Lower (Green) Arm Voltages (V): System Response to
Unbalanced Initial Conditions

62
The balancing loop acts as a disturbance for the overall energy loop, trying to reduce
the unbalance with the same charging current and thus slowing down the dynamic
response of voltages.
In order to reach the overall energy reference, control loop requires a high value of
differential current, which instantaneously charges both upper and lower capacitors.
The problem is mainly caused by the fast response of the overall energy loop: the
differential current required, in order to set to zero the error, charges both upper and
lower arm sub-modules, leading to a non-acceptable overshoot.
Thus an improvement in the control strategy is required, taking into account that the
differential current will charge anyway both upper and lower sub-modules.
In the event that one could ignore output variables behavior during the transient, it
would be easy to eliminate the overshoot: the already charged arm would be
bypassed, eluding the problem shown in previous simulations. But, as the tracking of
output variable references is imposed, no significant improvement can be obtained
from “low level” control.
Thus, the best solution is to work directly on the differential current: this approach
will permit to guarantee output variables to follow reference (with just a brief
transient) and reduce significantly overshoot in the energy response.
It is important to note that a significant overshoot reduction cannot be achieved
retuning the parameters of regulators; this because the small capacitors used in sub-
modules have a charging dynamic too fast for preventing completely overshoots by
acting on the converter regulators .
In this thesis, parameters will not be changed, but instead an improved control scheme
will be investigated and proposed. In order to prevent energy and voltage overshoots,
differential current is monitored and controlled. As it is not possible to directly control
this current, thus a feed-back-like structure is realized on the differential voltage loop
that influences overall energy loop.
If differential current overcomes a safe range, a proportional action acts in order to
decrease the differential voltage contribution and thus limit differential current. In
Fig. 41 the scheme representing the principle of operation is shown. Violet blocks are
used to compare differential current with an equivalent saturated current (saturation is
imposed to remain in a safe range of operation) and a proportional controller is then
imposed to the “error”.
Figs. 42 and 44 show the improvement of energy and arm voltages behavior: the
differential current control reduces arm voltages overshoots, and thus overall energy
overshoot. Actually, uncharged sub-modules reach voltage reference smoothly,
without overshoots: the already charged arm has however a small overshoot, due to
charging current. The behavior of the converter is now acceptable, and the overall
energy overshoot is less than 10%. Balancing of energy behavior remains almost the
same, compared to previous simulations and it is shown in Fig. 43.

63
Figure 41 - Improved Control Strategy

64
 6
x 10
2.2

2.1

1.9

1.8

1.7

1.6

1.5

1.4

1.3

1.2
0 1 2 3 4 5 6
Time
Time (s)
Figure 42 - Overall Energy (Joule): System response from an unbalanced initial
condition with the new regulator

5
x 10
1

-1

-2

-3

-4

-5

-6

-7

-8
0 1 2 3 4 5 6
Time
Time (s)
Figure 43 - Difference between Upper and Lower Arm Energy (Joule): System response
from an unbalanced initial condition with new regulator

65
 4
x 10
2.4

2.2

1.8

1.6

1.4

1.2

1
0 1 2 3 4 5 6
Time
Time (s)
Figure 44 - Upper (Blue) and Lower (Green) Arm Voltages (V): System response from
an unbalanced initial condition with the new regulator

66
4.5.4 - Simulation Results: LVRT

Last simulations shown in this thesis are related to Low Voltage Ride Through issue:
in order to simulate this operating condition, grid voltage is set to zero for 200 ms
as represented in Fig. 45.

8000

6000

4000

2000

-2000

-4000

-6000

-8000

2.8 2.9 3 3.1 3.2 3.3 3.4

Time
Time (s)
Figure 45 – Grid Voltage (V) during LVRT

The aim of this simulation is to see the energy behavior of the system; internal
transient and current ripples need a more precise and focused modeling approach, but
it is not in the scope of this thesis. What the simulation is supposed to proof is the
reliability and safety of energy control system in faulty operating conditions.
It is plausible to expect a transient before the system reaches a new “fail” steady-state:
after this condition, a new transient will lead the system back to the healthy operating
conditions.
Actually, the system behavior fulfills expectations: as shown in Fig. 46, the overall
energy is stable and has a small transient, leading to a steady-state condition.
Balancing control too, is able to maintain upper and lower arm voltages to the same
desired value, as illustrated in Fig. 47.
Fig. 48 and 49 show the waveforms of capacitor voltages and differential current,
respectively. The transient during LVRT is coherent with what expected. Differential
current mean value is zero during the fault as the output power drops to zero. The

67
alternate component instead remains almost unchanged, in order to maintain the
balance between upper and lower arm voltages.
6
x 10
2.06

2.04

2.02

1.98

1.96

1.94

1.92

1.9
0 1 2 3 4 5 6
Time(s)
Time
Figure 46 - Overall Energy (Joule): Behavior in LVRT condition
4
x 10
6

-2

-4

-6

-8
0 1 2 3 4 5 6
Time
Time (s)

Figure 47 - Difference between Upper and Lower Arm Energy (Joule): Behavior in
LVRT condition

68
 4
x 10
2.06

2.04

2.02

1.98

1.96

1.94

1.92
0 1 2 3 4 5 6
Time(s)
Time
Figure 48 – Upper (Blue) and Lower (Green) Arm Voltages (V): Behavior in LVRT
condition

500

400

300

200

100

-100

-200
0 1 2 3 4 5 6
Time
Time(s)

Figure 49 - Differential Current (A): Behavior in LVRT condition

69
70
Chapter V

5.1 – Motivation for a new modelling approach

From results obtained and showed in previous chapter, it possible to conclude that
linear control behaves properly in steady state condition and in certain dynamic
operation condition.
However, as underlined in Section 3.3, it is really hard to investigate advanced and
robust control strategy, due to high non linearity in model equations.
Even if the linear control strategy gives a satisfying response, parameters changing,
failure modes, fast dynamic requirements scenarios may introduce problems in the
stability of the system: thus, these possibilities call for a deepening in different control
approaches (for instance adaptive, robust, feedback linearization, optimal control etc).
In order to facilitate these studies, a new model approach here is proposed: the aim of
this modelling approach is to obtain “more linear” equations: even if it is not possible
to derive a completely linear model, it is desirable to apply advance control
techniques to a simpler set of equations.

5.2 – Modelling Principles and Equations

A new simplified schematic of the MMC leg is proposed and shown in Fig: 50: grid is
+EDC
not represented because not strictly necessary in this analysis, but must be introduced
to build a simulation model.

SMs
CP
P EP
VP
iP Zp

ZN iO
iN VO
VN SMs
N CN EN

Figure 50 - New Simplified MMC Leg Model

71
With reference to Fig 50, it is possible to write voltage loop and current equations
following Kirchhoff’s laws.

Defining Sp and Sn as modulation indexes, respectively of upper and lower arms, it is

possible to write:

It is also possible to define dynamic equations for equivalent capacitors voltages,

depending on modulation indexes:

It is possible to assume that output current reference is assigned: the problem of

controlling this current has a consequence in arm currents control.
These arm currents are supposed to maintain the overall energy to a desired value, to
balance the difference between upper and lower arm energy and to guarantee output
current to follow assigned set-point.
Thus, it is possible to write equations (53) and (54), in which the differential current is
introduced:

From (53) and (54), it is possible to calculate upper and lower arm current references:

72
The problem can be easily solved if differential current reference is assigned: as
mentioned before, this quantity has to be determined in order to maintain sub-module
capacitors charged and balanced.
Using equations (55) and (56) it is possible to calculate the arm current references,
then it is possible to create the current control loops as represented in Fig. 51.

Figure 51 - Currents Control Loops

From equations (49) and (50) it is then possible to calculate modulation indexes,
whose expressions are given in (57) and (58):

In order to find differential current reference, it is necessary to write and analyze

energy equations of MMC leg: instead of writing upper and lower energy equations,
total energy and unbalance energy equations are suggested.
Let’s consider WT and WD as total energy and difference energy between upper and
lower arms, respectively:

73
Introducing capacitor voltages and currents references, then equations (61) and (62)
are obtained:

It is possible to assume that voltage drop contribution created by currents on arm

impedance is negligible in comparison with capacitors voltage contribution: in order
to find a dynamic relationship between energy variations and variables available to
control MMC, equations (63) and (64) are introduced.

It is necessary to point out differential current contribution: thus, substituting upper

and lower arm currents with their reference expressions, i.e. equations (55) and (56),
yields to

( )

( )

If output current and voltage are periodical functions, with period T0 it is possible to
integrate equations (65) and (66), leading to

∫ ( ) ∫

74
 ∫ ∫ ( )

These equations show clearly differential current contribution in system energy

dynamic operation: total energy WT depends form differential current mean value,
WD instead depends form the alternate component of differential current having the
same frequency as output voltage.
These considerations can be summarized with the following scheme, Fig. 52.

Figure 52 – Energy Control Scheme Diagram

5.3 – Control Proposal

In this Section, control scheme previously described is explained extensively.
Compared with the control approach of Chapter III, used for simulation results in
Chapter IV, this new approach provides a cascade structure.
Two different loops are realized, an energy loop and a differential current loop:
energy loop provides a differential current reference, which is tracked by the current
control loop.
Energy loop is composed of overall energy control and balancing control.
Current loop is composed of upper current control and lower current control.
The complete control scheme is shown in Fig. 53.

75
Figure 53 - Control Scheme for new Modeling Approach

76
Following equations (46) and (47), current control loop is pretty easy to be tuned:
actually, between upper and lower arm voltages and arm currents only an R-L low
pass filter is present. Then, through equations (49) and (50), modulation indexes, real
system input, can be calculated.
Once control loop is tuned, it is possible to work on energy loop: using equations (65)
and (66), it is possible to control total energy and balancing dynamics. Thus, an
integrator is not necessary to set to zero total energy error and balancing error (i.e. an
integrator is intrinsically present in the system).
Control suggested and shown in Fig. 53 includes PD controller: this control structure
is again based on a frequency decoupling, where balancing loop is tuned to be slower
than total energy loop.
Simulation parameters chosen are the same used in the Mega-Watt system discussed
in Section 4.5.
In Figs. 54, 55 and 56, the simulation of a steady state condition is shown: the initial
transient is due to equation structure. A differential current derivative equation is not
present in the model, thus it is not possible to impose the initial condition.
It is however possible to see that after 2 s a steady state condition is reached without
overshoot or oscillation.
4
x 10 Ep
2.02

1.98

1.96

1.94
0 1 2 3 4 5 6

4
x 10 En
2.02

1.98

1.96

1.94
0 1 2 3 4 5 6
Time

Figure 54 - Upper and Lower Arm Capacitor Voltages

77
 Id_ref
1000

800

600

400

200

-200
0 1 2 3 4 5 6
Time

Figure 55 - Differential Current Transient (A)

Io
2000

1000

-1000

-2000
0 1 2 3 4 5 6

4
x 10 Vo
2

1.5

0.5

0
0 1 2 3 4 5 6
Time

Figure 56 - Output Current (A) and Output Voltage (V)

78
Further simulations, concerning dynamic behaviour of the system, will be carried out
in the near future: however, it was considered important to show a simple simulation
introducing the validity of the approach.
A comparison with model proposed in [8] (that has been used in the beginning of this
thesis to deepen system operation and control strategies) will also be developed: as an
introduction, next Section will briefly compare the equations of the two models in
order to propose future control possibilities.

5.4 – Future Works

Besides to basic control strategy described in previous paragraph, new modeling
approach gives a set of equations which are easier to be used in order to deepen
advanced control strategies.
It is possible to compare energy equations between the new proposed model and the
model described in [8]:

( )

( )

( )

( )

Neglecting notation differences, the high non-linearity of equations (34) and (35) is
smoothed considerably in equations (65) and (66).
Total energy in equations (34) is controlled through differential current, which is a
“square” non-linear input for the system (see Section3.3): in equation (65) instead, the
input is linear because the differential current is multiplying the DC voltage, which
can be considered constant.
A similar consideration may be done about balancing equations: equation (35) is
highly non-linear for input variables (differential voltage and differential current),
instead equation (66) has a component of the input which is still linear and can be
used to implement advanced control strategies.
Equations (34) and (35) give few chances to develop advanced control techniques,
due to their intrinsic non-linear structure.

79
Equations (65) and (66) instead, make it easier to define advanced control strategies
based on model inversion, feed-forward actions, and feed-back linearization.
All these possibilities will be investigated in near future and their functionality and
performances will be compared with standard linear controllers already implemented
and discussed in this work.

80
Chapter VI – Conclusions

The aim of this project was the analysis of a Modular Multilevel Converter (MMC)
for wind farm applications and the development of a control scheme to monitor the
energy behavior.
The analysis was based on the use of a simplified circuit, constituted by a single leg of
the converter, where all the modules in each arm were represented by a single variable
voltage source. The circuit model was derived as a system of differential equations,
used for analyzing both the steady state and dynamic behavior of the MMC, from
voltages and thus energy point of view.
Preliminary analysis was carried out following the modeling approach proposed in
[8]. Using this model, the system behavior has been studied and a possible energy
control scheme has been developed. The numerical simulations have shown that the
system is operating correctly under steady-state and transient operating conditions,
except for initial conditions with unbalanced capacitor voltages. To reduce the large
voltage overshoots due to this unbalanced initial condition, an improved control
scheme has been developed.
In Chapter V, a new modeling approach was proposed, aiming to define a set of
equations more suitable for advanced control implementation.
With the new modeling approach a significant contribution to control issues will be
possible: in the near future, a comparison between the new modeling approach and the
one proposed in [8] will be investigated.
New control structures will also be developed, in order to accomplish all dynamic
requirements required by the particular application, i.e. to comply with the new grid
code standards.

81
82
REFERENCES
[1] J. Rodriguez, J. S. Lai, and F. Z. Peng, “ Multilevel Inverters: a survey of topologies, controls, and
applications. “ IEEE Trans. Ind. Electronics, vol. 49, no. 4, pp. 724-738, August 2002.

[2] A. Lesnicar, and R. Marquardt, “An Innovative Modular Multilevel Converter Topology Suitable
for a Wide Power Range”, IEEE PowerTech Conference, Bologna, Italy, June 23-26, 2003.

[3] A. Lesnicar, and R. Marquardt, “A new modular voltage source inverter topology”, EPE 2003,
Toulouse, France, September 2-4, 2003.

[4] R. Marquardt, and A. Lesnicar, “New Concept for High Voltage - Modular Multilevel Converter”,
IEEE PESC 2004, Aachen, Germany, June 2004.

[5] M. Glinka and R. Marquardt, “A New AC/AC Multilevel Converter Family”, IEEE Transactions
on Industrial Electronics, vol. 52, no. 3, June 2005.

[6] B. Gemmel, J. Dorn, D. Retzmann, and D. Soerangr, “Prospects of multilevel VSC technologies for
power transmission”, in Conf. Rec. IEEE-TDCE 2008, pp. 1-6.

[7] M. Hagiwara, H. Akagi “PWM Control and Experiment of Modular Multilevel Converters”, IEEE
PESC 2008, Rhodes, Greece, June 2008.

[8] A. Antonopoulos, L. Angquist, and H.-P. Nee, “On dynamics and voltage control of the modular
multilevel converter,” European Power Electronics Conference (EPE), Barcelona, Spain,
September 8-10, 2009.

[9] G.Bergna, M.Boyra, J.H.Vivas, “Evaluation and Proposal of MMC-HVDC Control Strategies
under Transient and Steady State Conditions”

[10] P.Munch, S.Liu, M.Dommaschk “Modeling and Current Control of Modular Multilevel
Converters Considering Actuator and Sensor Delays”

[11] M.Hagiwara, R.Maeda, H.Akagi, “Control and Analysis of the Modular Multilevel Cascade
Converter Based on Double-Star Chopper-Cells (MMCC-DSCC)”

[12] W.Li, L.A.Regoire, J.Bèlanger OPAL-RT Technologies (CAN), “Control and Performance of a
Modular Multilevel Converter System”

[13] S. Rohner, J. Weber and S. Bernet “Continuous Model of Modular Multilevel Converter with
Experimental Verification”, Dresden University of Technology

83