978-1-4799-5776-7/14/$31.00 (c)2014 IEEE

Alternate Arm Converter Operation

of the Modular Multilevel Converter

M.M.C. Merlin, P.D. Judge, T.C. Green, P.D. Mitcheson

Imperial College London

London, UK

michael.merlin@imperial.ac.uk

F. Moreno, K. Dyke

Alstom Grid

Stafford, UK

francisco.moreno@alstom.com

Abstract—A new operating mode of the Modular Multilevel

Converter (MMC) using modified arm current waveforms

inspired from the working principle of the Alternate Arm

Converter (AAC) is presented in this paper. A reduction in the

cell voltage deviation is observed at power factors close to unity

at the cost of an increase in power losses, especially when

reactive power is required. This gain in voltage margin is then

used in further optimizations of the MMC performance, mainly

focusing on either increasing the number of redundant cells or

improving the overall power efficiency of the converter.

I. INTRODUCTION

The Modular Multilevel Converter (MMC) [1], illustrated in

Fig. 1, was the first modular Voltage Source Converter (VSC)

to provide both low switching losses and low AC current

distortion; these features contribute to the higher power

efficiency and the significant reduction in the size of the AC

filters of recent VSC power stations [2]. A number of MMC

converter stations are now in operation and more are planned

in the next couple of years with power ratings ever increasing.

Over the last few years, a new family of hybrid topologies [3-

6], such as the Alternate Arm Converter [4,5] (AAC), drawn

in Fig. 2, have emerged. These topologies essentially

combine both the 2-level VSC and the MMC together in

order to improve on some characteristics such as smaller

overall volume, better power efficiency and, in some

topologies, DC-side fault blocking capability. It has also been

shown [7] that the AAC has a better temperature distribution

between the IGBT modules of a cell in comparison to the

MMC. Extensive studies have also presented different

modelling approaches of the electical dynamics of these

multilevel converters and it has been shown that reduced

dynamic models have an acceptable level of accuracy while

offering an appreciable boost in computing speed [8-11]. The

AAC has a relatively similar electrical topology to the MMC

with the main difference being the presence of director

switches, i.e. series-IGBTs, in its arms in series with the stack

of cells. The switching state of these director switches

determines which arms are both generating the converter

voltage waveform and carrying the AC current to the

respective DC terminal. This approach has proven [4,5] to be

both power and volume effective, resulting in (i) the number

of cells be significantly reduced, e.g. up to half the number of

cells compared to the MMC, (ii) the average cell voltage

deviation is lower allowing smaller capacitors in the cells and

(iii) the use of full H-bridge cells instead of half-bridge cells,

thus the AAC is able to block DC-side faults without

compromising its overall power efficiency. In the case of the

MMC, it was observed rapidly after the topology was

proposed that unplanned circulating currents (mainly second

harmonics) were running between the legs [12]. This issue

was later solved by implementing additional feedback loops

in charge of monitoring the arm currents [13]. Besides, it has

also been observed that this circulating current can be used to

reduce the cell voltage deviation of the cell capacitors [12] or

by adding a 2nd harmonic filter at the midpoint of the arm

inductors [14].

This paper presents an original arm current waveform for the

in the MMC, inspired by the working mechanism of the

AAC. This study focuses on the effects of changing the arm

current waveforms which only implies an update of the

control system (software) but not the converter itself

(hardware) which thus stays the same. This implies that (i)

both the AC and DC quantities (e.g. power, voltage and

current magnitudes) remain unchanged, (ii) the cells are still

of the half-bridge type and (iii) the passive components (e.g.

cell capacitors and arm inductors) keep the same values.

Furthermore only the top level part of the controller is

affected (e.g. current control) but the low-level control stays

the same thus this does not require changes in either the

wiring or the communication with the cells. Finally the

benefits of this update both in terms of cell capacitor voltage

deviation and power efficiency will be explored and margins

for further optimizations are discussed. Finally, modular

VSC topologies such as the MMC and the AAC can also be

connected in front-to-front arrangements (meaning that the

AC sides are facing each others) in order to form a DC/DC

converter such as in [15]. Therefore the current modulation

concept introduced in this paper can potentially also be used

in such DC/DC topologies with equivalent pros and cons as

discussed in this paper.

This study has been financed by the EPSRC under the UK Power

Electronics Centre – Converter Theme grant (EP/K035096/1)

Figure 1 - Half-bridge MMC topology

Figure 2 - AAC topology

II. ARM CURRENT WAVEFORMS

In its classic operating mode, the MMC distributes equally (i) the AC current between its upper and lower arms and (ii) the DC current between the three different legs. The complete set of equations describing the arm currents is given in (1)-(6). The signs present in these equations depends on the direction of the currents, using what is illustrated in Fig 1 and 2.

IA+ =

IA

2+

𝐼𝐷𝐶

3 (1)

IA− = −

IA

2+

𝐼𝐷𝐶

3 (2)

IB+ =

IB

2+

𝐼𝐷𝐶

3 (3)

IB− = −

IB

2+

𝐼𝐷𝐶

3 (4)

IC+ =

IC

2+

𝐼𝐷𝐶

3 (5)

IC− = −

IC

2+

𝐼𝐷𝐶

3 (6)

The AAC distributes its AC currents in a different pattern as only one arm in each leg carries the full AC current while the other has its director switch opened, blocking any current from flowing through the arm. Using the variables αA,B,C ∈ {0,1} which represent whether the AC current of a particular leg is passing either through the top or bottom arm, the set of equations describing the arm currents in AAC mode are given in (7)-(12):

IA+ = 𝛼𝐴 𝐼𝐴 −

𝐼𝐹

3 (7)

IA− = −(1 − 𝛼𝐴)𝐼𝐴 −

𝐼𝐹

3 (8)

IB+ = 𝛼𝐵 𝐼𝐵 −

𝐼𝐹

3 (9)

IB− = −(1 − 𝛼𝐵)𝐼𝐵 −

𝐼𝐹

3 (10)

IC+ = 𝛼𝐶 𝐼𝐶 −

𝐼𝐹

3 (11)

IC− = −(1 − 𝛼𝐶)𝐼𝐶 −

𝐼𝐹

3 (12)

However the AAC inherently generates a 6-pulse ripple on top of the DC current waveform as a consequence of the alternating nature of the arm conduction periods. Since the objective of this study is to migrate an already existing MMC converter to an AAC mode of operation without having to change the hardware but only by updating the control system (i.e. software), installing a DC filter is out of question. To resolve this issue, an additional active filtering current is added to the arm currents in order to keep the DC current smooth. This filtering current IF (13) is obtained by calculating the ripple component of the DC current waveform resulting from AAC operation in order to suppress it by adding a circulating current of opposing sign through all three legs.

𝐼𝐹 = 𝐼𝐴+ + 𝐼𝐵

+ + 𝐼𝐶+ − 𝐼𝐷𝐶

= 𝐼𝐴− + 𝐼𝐵

− + 𝐼𝐶− − 𝐼𝐷𝐶

= 𝛼𝐴 𝐼𝐴 + 𝛼𝐵 𝐼𝐵 + 𝛼𝐶 𝐼𝐶 − 𝐼𝐷𝐶 (13)

The addition of this active filtering current alters the original nature of the AAC mode as the arms will now continuously run a current through them as opposed to for only one half of the cycle and no current during the other half of the cycle because the director switches would be closed. However the magnitudes of the arm currents will be small for a large proportion of the time, while the arm is not carrying the main AC current, compared to the MMC mode of operation, as shown in the simulation section below. Finally, only the arm current waveform will be different because the currents seen at both the AC and DC terminals of the converter will be the same as under the normal MMC mode of operation.

III. SIMULATION

A. Simulation Model

In order to assess the benefits of the AAC mode of operation, a 120 MW MMC converter has been simulated using Simulink® and the Artemis® toolbox in order to simulate a reasonably large number of half-bridge cells. The characteristics of this simulation model are listed in Table 1. In order to better match realistic MMC converter, triplen harmonic voltage injection has been used (about 15% of the

fundamental magnitude) in order to shape the converter voltage waveform into an almost trapezoidal waveform and to push the power rating to its theoretical maximum while still using only half-bridge cells. Furthermore, a power loss analysis has also been performed by post-processing the voltage and current waveforms in each cells using the power loss data from the datasheet of the 3.3 kV 1.2 kA HiPak IGBT device 5SNA 1200E330100 [16] using the method described in [7, 17] . In the later part of the simulation section, the steady state junction temperature of the different semiconductor devices in a cell (e.g. top and bottom IGBT and diode modules) by applying the individual power loss figure of each device into a power-thermal model derived in ANSIS and verified in [17] and assuming a coolant temperature of 60ºC shared by all the semiconductor devices.

Table 1 - Characteristics of the simulated MMC model

Rated power 120 MW

DC bus ±50 kV

AC line 55 kV

AC frequency 50 Hz

Triplen harmonic voltage 15%

Number of cells per stack 56

Cell capacitor 8 mF

Phase inductor 8 mH

Arm inductor 6 mH

B. Arm current waveforms

The arm current waveforms resulting from this mode of

operation are observed in this section. Fig. 3 shows the top

arm current waveform in both MMC and AAC operating

modes (respectively the red and green curves) under unity

power factor rectifier mode. The third curve (blue) is the

result of the difference between the two waveforms and can

be interpreted as the equivalent circulating current which can

be injected in the converter to move from the MMC mode to

the AAC mode of operation. It can be observed that at unity

power factor a large amount of the fundamental cycle is spent

with a small amount of current magnitude in AAC mode as

opposed to the MMC mode. As more reactive power is

involved in the conversion process as shown in Fig 4 (same

amount of active and reactive power) and in Fig. 5 (reactive

power only), the arm current waveform becomes more and

more distorted in AAC mode with still a significant portion

of the cycle used by low magnitude current but at the expense

of an increasing peak values to attain twice the value in the

MMC mode. The power losses have computed for different

operating points and the results listed in Table 2 with the total

power loss values plotted in accordance to the angular value

of their respective operating points in Fig. 6.

Figure 3 - Arm A+ current waveform in MMC (red), AAC (green)

modes and the resulting circulating current (blue) with active power

only

Figure 4 - Arm A+ current waveform in MMC (red), AAC (green)

modes and the resulting circulating current (blue) with both active

and reactive powers

Figure 5 - Arm A+ current waveform in MMC (red), AAC (green)

modes and the resulting circulating current (blue) with reactive

power only

Table 2 – Power losses of the MMC model depending on the mode of operation and operating point relative to the apparent power

Active Power Reactive Power

1 pu 0 pu

0.7 pu 0.7 pu

0 pu 1 pu

-0.7 pu 0.7 pu

-1 pu 0 pu

-0.7 pu -0.7 pu

0 pu -1 pu

0.7 pu -0.7 pu

MMC mode

Conduction losses

0.50% 0.44% 0.36% 0.37% 0.38% 0.37% 0.35% 0.45%

Switching losses

0.15% 0.14% 0.13% 0.13% 0.15% 0.15% 0.15% 0.16%

Total losses

0.65% 0.58% 0.49% 0.51% 0.53% 0.52% 0.50% 0.60%

AAC mode

Conduction losses

0.53% 0.56% 0.51% 0.43% 0.38% 0.40% 0.49% 0.53%

Switching losses

0.16% 0.22% 0.24% 0.20% 0.17% 0.26% 0.30% 0.32%

Total losses

0.69% 0.78% 0.75% 0.63% 0.55% 0.66% 0.79% 0.85%

Figure 6 - Power losses as percent of the total apparent power in

MMC and AAC modes for different operating points

C. Observation at unity power factor

The previous set of results indicates that this new AAC mode

is only potentially attractive for unity power factor only as

both the arm current peak values and the power losses are

increase dramatically when reactive power is involved in the

conversion process. The next part of this paper assumes that

only these two operating points (inverter and rectifier active

power only) are considered with the results focusing on the

rectifier mode mainly since the inverter mode is merely an

opposite phase angle version of the former.

Figure 7 - Arm current waveforms in MMC mode

Figure 8 - Arm current waveforms in AAC mode

Fig. 7 and 8 show the arm current waveforms respectively in MMC and AAC modes of operation. On one hand, the MMC mode results in the arms continuously conducting a large amount of current as opposed to the AAC mode where a significant part of the cycle is spent with only a small amount

of current (around -150 A). This remaining low magnitude current mainly consists of the filtering current IF (13). On the other hand, the peak value of the arm current currents is lower in the MMC mode compared to the AAC mode (respectively 1.3 kA and 1.6 kA thus 23% higher in the AAC case).

Figure 9 – Upper cell voltage waveforms in MMC mode at normal

PWM frequency (Fpwm = 169 Hz)

Figure 10 – Lower cell voltage waveforms in MMC mode at normal

PWM frequency (Fpwm = 169 Hz)

The other observable difference between these two modes of operation is the voltage deviation of the cell capacitors. Fig 9 and 11 present both the average and individual cell voltages in one arm for respectively the MMC and AAC modes at the same phase-shifted PWM [18] at the same frequency (169 Hz). The AAC mode exhibits a lower voltage deviation, both in average and individual voltages, due to the way the voltage and current waveforms of its arms combine to create the energy exchange. Furthermore, the waveforms of the upper and lower stack voltages are similar in each modes of operation (Fig. 9 and 10 for the MMC mode and Fig. 11 and 12 for the AAC mode), with the lower stack voltage waveform being essentially a time-shifted version of the upper stack voltage waveform. The individual cell voltages are more evenly distributed around the average value in the AAC mode of operation compared to the MMC mode. In the next section of this paper, only the upper stack voltage will considered. In terms of power efficiency, the two modes of operation

generate approximately the same amount of power losses at unity power factor as noted previously with the MMC mode performing slightly better in switching losses while the AAC mode wins on conduction losses. The simulation results are summarized in Table 3.

Figure 11 – Upper cell voltage waveforms in AAC mode at normal

PWM frequency (Fpwm = 169 Hz)

Figure 12 – Lower cell voltage waveforms in AAC mode at normal

PWM frequency (Fpwm = 169 Hz)

IV. OPTIMIZATION SCENARIOS

A. Reduced PWM frequency

Since the AAC mode of operation improves the cell voltage deviation, two possible optimizations can be done. The first possible optimization consists of reducing the PWM frequency. This will result in (i) higher differences between the cell voltages but (ii) cell voltages still approximately within the same envelope as in the MMC mode while (iii) the switching losses will decrease. Fig. 13 and Table 3 (second “AAC” column) show the results for a reduced PWM frequency at 119 Hz. As predicted, the average cell voltage is unaffected but the individual cell voltages are further distributed around the average cell value but still approximately within the limits of the MMC mode of operation. The conduction losses are unchanged because the current waveforms are unchanged but the switching losses decreased because the IGBTs are not switching as often, leading to a marginally better power efficiency figure.

Table 3 - Simulation results in rectifier mode at unity power factor

Mode of operation MMC AAC AAC AAC

Average cell voltage 1.800 kV 1.800 kV 1.800 kV 1.880 kV

PWM carrier frequency 169 Hz 169 Hz 119 Hz 169 Hz

Average peak-to-peak cell voltage deviation

252 V 172 V 172 V 166 V

Maximum individual cell voltage 1,980 V 1,954 V 1,942 V 2,003 V

Minimum individual cell voltage 1,660 V 1,696 V 1,673 V 1,767 V

Arm current (Maximum value) 1,323 A 1,643 A 1,643 A 1,643 A

Arm current (Minimum value) -503 A -244 A -244 A -244 A

Arm current (Average value) 402 A 402 A 402 A 402 A

Arm current (RMS value) 753 A 803 A 803 A 803 A

Conduction losses 468 kW 459 kW 459 kW 459 kW

Switching losses 180 kW 202 kW 156 kW 209 kW

Total power losses 648 kW 661 kW 615 kW 668 kW

Power efficiency (semiconductor only) 99.46% 99.45% 99.49% 99.44%

Temperature IGBT 1 (steady state) 75ºC 74ºC 72ºC 73ºC

Temperature Diode 1 (steady state) 74ºC 72ºC 70ºC 72ºC

Temperature IGBT 2 (steady state) 74ºC 75ºC 75ºC 76ºC

Temperature Diode 2 (steady state) 100ºC 104ºC 103ºC 104ºC

Number of extra redundant cells 0 0 0 2.8 (5%)

Figure 13 – Upper cell voltage waveforms in AAC mode with

reduced PWM frequency (Fpwm = 119 Hz)

B. Increased average cell voltage

The second possible optimization is to use the maximum cell voltage headroom made available by the reduced cell voltage deviation. This optimization will result in a small number of cells (around 5% in this case) being made redundant, thus increasing the overall reliability of the converter station in case of cell failures. Fig. 14 shows that the cells can be charged up to the new nominal value of 1,880 V with the highest peak value for the individual cell voltages being not much higher than in the MMC case.

Figure 14 – Upper cell voltage waveforms in AAC mode with higher

nominal cell voltages (Vcell = 1,880 V)

Table 3 shows that the switching losses went up proportionally to the higher nominal cell voltage. However, it could be possible to reduce further the power losses, e.g. by bypassing electrically the now-redundant cells. The peak-to-peak voltage deviation is lower than in the other AAC mode cases because of the quadratic nature of the relationship between the cell voltage deviation and the energy deviation. The latter is still unchanged but the higher nominal cell voltage means that the energy deviation will happen at a higher part of the quadratic curve, hence the slightly smaller resulting voltage deviation.

V. CONCLUSION

The AAC mode of operation of an existing MMC station has been presented in this paper. This new mode of operation does not require any hardware modification of the converter station but rather an update of the control system, hence mainly software. Generally the power losses increase in the AAC mode of operation compared to the MMC mode but are approximately the same when operating close to unity power. Given this fact, the study focused only on active power conversion operating points. In this condition, the AAC-mode still exhibits higher peak arm current combined with a significant part of the cycle at low current magnitudes but, most importantly, smaller cell voltage deviations (e.g. 30% lower) compared to the classic MMC mode are observed.

This last fact can be used to further optimize the performance of the MMC converter station in two different ways. First, reducing the switching frequency of the PWM signal will result in a substantial decrease in the switching losses, hence a higher power efficiency figure for the converter station. This leads also to an increase of the individual cell voltage deviations but those are still contained within the same limits set during the previous MMC mode. Second, increasing slightly the nominal voltage of the cells results in having some cells being made redundant (approximately 5% of the cells in each stack) since the stacks are able to produce more voltage than normally required. Despite the higher nominal voltage, the maximum peak value hit by the cell voltages is still the same maximum value set during the previous MMC mode of operation. From here, either (i) these redundant cells are electrically bypassed in order to further improve the power efficiency of the converter station, or (ii) they can be used as back-up cells in case of other cell failures, thus further improving the overall reliability of the MMC.

REFERENCES

[1] A. Lesnicar and R. Marquardt, “An innovative modular multilevel converter topology suitable for a wide power range,” in PowerTech Conference Proceedings, 2003 IEEE Bologna, vol. 3, jun 2003

[2] Sellick, R. L., and M. Åkerberg. "Comparison of HVDC Light (VSC) and HVDC Classic (LCC) Site Aspects, for a 500MW 400kV HVDC Transmission Scheme." ACDC 2012, Birmingham UK.

[3] D. R. Trainer, C. C. Davidson, C. D. M. Oates, N. M. MacLeod, D. R. Critchley, and R. W. Crookes, “A new hybrid voltage-sourced converter for HVDC power transmission,” CIGRE Paris Sess. 2010, 2010.

[4] M. Merlin, T. Green, P. Mitcheson, D. Trainer, D. Critchley, and R. Crookes, “A new hybrid multi-level voltage source converter with dc fault blocking capability,” in AC and DC Power Transmission, 2010. ACDC. 9th IET International Conference on, Oct 2010.

[5] Merlin, M.M.C.; Green, T.C.; Mitcheson, P.D.; Trainer, D.R.; Critchley, R.; Crookes, W.; Hassan, F., "The Alternate Arm Converter: A New Hybrid Multilevel Converter With DC-Fault Blocking Capability," Power Delivery, IEEE Transactions on , vol.29, no.1, pp.310-317, Feb. 2014.

[6] Adam, Grain Philip, et al. "New breed of network fault-tolerant voltage-source-converter HVDC transmission system." Power Systems, IEEE Transactions on 28.1 (2013): 335-346.

[7] Judge, P.D.; Merlin, M.M.C.; Mitcheson, P.D.; Green, T.C., "Power loss and thermal characterization of IGBT modules in the Alternate Arm converter," Energy Conversion Congress and Exposition (ECCE), 2013 IEEE , vol., no., pp.1725-1731, 15-19 Sept. 2013

[8] Saad, H.; Dennetiere, S.; Mahseredjian, J.; Delarue, P.; Guillaud, X.; Peralta, J.; Nguefeu, S., "Modular Multilevel Converter Models for Electromagnetic Transients," Power Delivery, IEEE Transactions on , vol.29, no.3, pp.1481-1489, June 2014.

[9] Peralta, Jaime, et al. "Detailed and averaged models for a 401-level mmc–hvdc system." Power Delivery, IEEE Transactions on 27.3 (2012): 1501-1508.

[10] Beddard, A.; Barnes, M.; Preece, R., "Comparison of Detailed Modeling Techniques for MMC Employed on VSC-HVDC Schemes," Power Delivery, IEEE Transactions on , vol.PP, no.99, pp.1,11

[11] Caitríona E. Sheridan; Michaël M. C. Merlin; Timothy C. Green, “Reduced Dynamic Model of the Alternate Arm Converter”, COMPEL 2014, Santander, June 2014

[12] Wang, Kui, et al. "Voltage balancing and fluctuation-suppression methods of floating capacitors in a new modular multilevel converter." Industrial Electronics, IEEE Transactions on 60.5 (2013): 1943-1954.

[13] Antonopoulos, Antonios, Lennart Angquist, and H-P. Nee. "On dynamics and voltage control of the modular multilevel converter." Power Electronics and Applications, 2009. EPE'09. 13th European Conference on. IEEE, 2009.

[14] Jacobson, Bjorn, et al. "VSC-HVDC transmission with cascaded two-level converters." Cigré session. 2010.

[15] Luth, T.; Merlin, M.M.C.; Green, T.C.; Hassan, F.; Barker, C.D., "High-Frequency Operation of a DC/AC/DC System for HVDC Applications," Power Electronics, IEEE Transactions on , vol.29, no.8, pp.4107-4115, Aug. 2014

[16] ABB HiPak IGBT Module 5SNA 1200E330100. Datasheet available online. [Online]. Available: http://www.abb.com/semiconductors

[17] U. Drofenik, D. Cottet, A. Müsing, J. M. Meyer, and J. W. Kolar, “Modelling the thermal coupling between internal power semiconductor dies of a water-cooled 3300V/1200A HiPak IGBT module,” in Proceedings of Power Conversion and Intelligent Motion Conference, 2007

[18] Tu, Qingrui, Zheng Xu, and Lie Xu. "Reduced switching-frequency modulation and circulating current suppression for modular multilevel converters." Power Delivery, IEEE Transactions on 26.3 (2011): 2009-2017.