Anda di halaman 1dari 8

Analysis and Calculation of Zero-Sequence Voltage

Considering Neutral-Point Potential Balancing in


Three-Level NPC Converters
WANG Chenchen, Member, IEEE, and LI Yongdong.

Abstract—The neutral-point (NP) potential of the three-level Correspondingly, if carrier-based PWM (CPWM) is used,
neutral-point-clamped converters is needed to maintain the control of the NP potential can be considered as the
balancing. Zero-sequence voltage is the only freedom degree problem of identifying the zero-sequence voltage. The zero-
when carrier-based PWM is utilized. Appropriate zero-sequence sequence voltage injected into the reference voltages does not
voltage should be identified to control the NP potential. The change the output line voltages, but influences the switching
relationship between the neutral current and injected zero- states and of course the NP potential. The NP potential
sequence voltage is studied comprehensively, and two balancing variation caused by the injected zero-sequence voltage was
algorithms of NP potential respectively adopting searching- studied, and some algorithms to keep the NP potential
optimization and interpolation methods are presented. The
balancing by injecting the appropriate zero-sequence voltage
theoretical optimum zero-sequence voltage for controlling NP
potential can be obtained by the latter proposed algorithm.
were presented in [11-15]. The low frequency voltage
Simulation and experimental results are shown to verify the oscillations in the neutral point were also studied, and the
validity and practicability of the proposed algorithms. solution was presented in [16-18]. But higher switching
frequency and extra switching losses of converter are
Keywords-NPC; NP potential; balancing algorithm; zero- unavoidable.
sequence voltage. It’s known that, the zero-sequence voltage is the only
freedom degree which can be adjusted in the process of
I. INTRODUCTION modulation. The relationship between SVPWM and CPWM
Three-level Neutral-Point-Clamped (NPC) converters[1] has been studied, and the identity of them has been obtained[19,
have been widely applied in the medium-voltage power 20]. Essentially, selecting redundant switching states and
conversion[2-4]. One of the essential problems of the NPC adjusting dwell time in SVPWM is the process of deciding the
converters is that how to keep the voltage of dc-link capacitors zero-sequence voltage. So the zero-sequence voltage should be
balancing, in other words, keep the neutral-point (NP) potential taken focus on when the problem of NP potential balancing is
stable and suppress the ripple[5]. If the NP potential is not considered. It is noticed that no matter which modulation
controlled effectively, the output voltage of the converter method is adopted, the neutral current is an important variable.
would deviate from the reference value, moreover the devices Once the neutral current is known, the NP potential variation
and equipments might be damaged. can be obtained. On the other hand, once the zero-sequence
voltage is identified, the switching state is determined. And
The control strategies of NP potential that have appeared in according to the three-phase currents and switching state, the
the literatures can be summarized according to the pulse width neutral current can be calculated. But the relationship between
modulation (PWM) algorithm utilized. If space vector PWM the neutral current and the zero-sequence voltage is not linear
(SVPWM) is used, the voltage vectors can be classified into and continuous, which causes that the algorithm is not direct
four categories by their magnitude: zero vectors, small vectors, and facility.
middle vectors and large vectors. Then the relationship
between NP potential and each switching state vector can be The relationship between neutral current and injected zero-
analyzed. It is known that the zero vectors and large vectors sequence voltage in three level NPC converters is studied in
have no effect on NP potential, but the middle vectors and this paper comprehensively. A searching-optimization
small vectors have effect on it. It’s noticed that there are two algorithm to balance the NP potential is presented, and an
switching states (positive and negative) that have reverse action interpolation method to calculate the exact zero-sequence
(charging or discharging) on NP potential for one small vector. voltage for NP potential balancing is proposed simultaneously,
Therefore the main task is to adjust the dwell time between the which is called precise calculation algorithm in this paper. The
duplicate switching states of small vectors[6-10]. two presented methods are verified through the simulation and
experimental results.
Manuscript received March 24, 2009, revised May 10, 2009; accepted for
publication May 15, 2009.
Copyright © 2009 IEEE. Personal use of this material is permitted. However, II. RELATIONSHIP BETWEEN NEUTRAL CURRENT AND ZERO-
permission to use this material for any other purposes must be obtained from SEQUENCE VOLTAGE
the IEEE by sending a request to pubs-permissions@ieee.org.
WANG Chenchen is with the School of Electrical Engineering, Beijing Fig.1 is a schematic diagram of a three-level NPC inverter.
Jiaotong University, 100044 Beijing, China (e-mail: chchwang@ bjtu.edu.cn). In order to simplify the analysis, several assumptions are made
LI Yongdong is with Department of Electrical Engineering, Tsinghua here:
University, 100084 Beijing, China. (e-mail: liyd@mail.tsinghua.edu.cn).
iCH

uH

ia
io
ic
va ,b ,c
ib
iCL

uL

Fig 2. Average phase voltage and time ratio of the neutral point output

Fig 1. Three-level NPC inverter I m : amplitude of the phase current;


 The dc-link voltage is constant. Then the influence of  : power factor angle.
the dc-link voltage on the NP potential can be eliminated.
Generally, only a couple of switching on and off in a
 The upper capacitor and the lower capacitor in the dc-
modulation period for a phase leg. For a three-level NPC
link have the same capacitance and characteristic, i.e.
converter, one phase current makes contribution to the neutral
CH  C L  C . current only when this phase is connected to the neutral point,
 The inverter is connected to a three phase balanced that is to say, when this phase outputs 0-level voltage. Fig.2
load. The three-phase voltages and currents are purely shows the relationship between an average phase voltage and a
sinusoidal and symmetrical. time ratio of the neutral point output[11]. Furthermore, it is
known that in a three-leg three-wire (3L3W) system,
 The modulation frequency is higher enough than the ia  ib  ic  0 . So the average neutral current in a modulation
output fundamental frequency, and the phase currents can be period can be obtained:
considered as constant in a modulation period.
If Vdc / 2 ( Vdc is defined as the dc-link voltage) is selected io   1  vx   ix    v x  ix (4)
x  a ,b , c x  a ,b , c
as the base value, the three phase positive-sequence reference
voltages can be normalized as: And the NP potential variation can be easily calculated by
the following equation:
va 0  m  (2 / 3)  cos 
1
2C 
vb 0  m  (2 / 3)  cos(  2 / 3) (1) uL   io dt (5)

vc 0  m  (2 / 3)  cos(  2 / 3)
For convenience of analysis, vmax0 / vmax , vmid0 / vmid ,
where vmin0 / vmin are defined as the maximum, medium, minimum
m : modulation index, and 0  m  1 ; (The maximum value of the reference voltages without/with injected zero-
modulation index can achieve 1 in linear modulation area by sequence voltage, and iv max , ivmid , iv min as the current of the
SVM or by carried-based PWM strategies that make use of corresponding phase. The injected zero-sequence voltage will
proper zero-sequence voltage) not modify the relation of magnitude, so
 : phase angle, and   t , where  is the angular vmax,mid,min  vmax0,mid0,min0  v0 (6)
frequency.
In this paper, the phase voltage is set as 1  vx  1 ,
If the zero-sequence voltage ( v0 ) is injected into the three
x  [a, b, c] , so the span of the zero-sequence voltage that can
phase reference voltages, the actual reference voltages can be
given by: be injected is expressed as:

va  va 0  v0 v0_min0  v0  v0_max (7)


vb  vb 0  v0 (2) where
vc  vc 0  v0
v0 _ min  1  vmin0 , v0 _ max  1  vmax0
According to the aforementioned assumption, the three-
phase currents of the converter can be given by are respectively defined as the smallest and biggest zero-
sequence voltages can be injected.
ia  I m cos(   )
In (4), it can be seen that the average neutral current is
ib  I m cos(  2 / 3   ) (3) related with the phase currents and the absolute value of the
ic  I m cos(  2 / 3   ) reference voltages. It seems like that the zero-sequence voltage
which can keep the NP potential balancing could be
where straightforward calculated. The only problem need to be
resolved is the polarity of phase voltage. [11] simply dealt with
b
it by considering the polarity in different section, and gave the
zero-sequence voltage forcing the neutral current to be zero in 020 120 220
section I as:
1
cos   cos(2   )
m2 021 010 110 210
v0  (8)
2 sin(   ) 121 221

But it’s noted that the polarity of phase voltage added the 000
zero-sequence voltage possibly reverses. Correctly, the injected 022 011 100 200
111
zero-sequence voltage should be expressed as[13]: 122 211 a
222
io   Ssgna  va 0  ia  Ssgnb  vb 0  ib  Ssgnc  vc 0  ic 
v0  (9) 001 101
Ssgna  ia  Ssgnb  ib  Ssgnc  ic 012 112 212 201

where io which can be obtained in terms of (9) is the expected


neutral current that keep the NP potential balancing. It’s
attended that here: 002 102 202
c
T T
[ Ssgna , Ssgnb , Ssgnc ]  [sgn(va ),sgn(vb ),sgn(vc )]

Unfortunately, sgn(vx ) is not certain when different zero- Fig 3. Three level space vector digram
sequence voltages are injected. In order to solve this problem, a the sign of vmid0 is not determined. But whatever the sign is,
test-verify-revise algorithm for NP potential balancing was
presented in [13]. Firstly, the polarity of phase voltage with the relationship between the neutral current and zero-sequence
injected zero-sequence voltage was assumed to be same as the voltage v0 can be derived as follows according to the value of
polarity of positive-sequence voltage, then the prediction result v0 :
v0test should be substituted into (7) and the validity of the
assumption on polarity was verified. If the result didn’t accord (1). v0  vmin0 , then the three phase reference voltages are
with the assumption, the sigh of vmid would be revised and all nonnegative. The average neutral current is:
substituted into (9) to obtain the new result. The change of the io  vmax  iv max  vmid  ivmid  vmin  iv min
other phase voltage was not considered and the trouble of the (10)
polarity caused by injected zero-sequence voltage was settled  vmax 0  iv max  vmid0  ivmid  vmin 0  iv min  io'
indirectly.
(2). vmid0  v0  vmin0 , then vmin  0 , vmid  0 ,
III. FURTHER ANALYSIS AND THE PROPOSED NP POTENTIAL vmax  0 . The average neutral current is:
BALANCING ALGORITHMS
io  vmax  iv max  vmid  ivmid  vmin  iv min
Actually, the most comprehensive method is to traverse all (11)
the zero-sequence voltages that can be injected, and choose the  vmax 0  iv max  vmid0  ivmid  vmin 0  iv min  2  iv min  v0
most helpful one for the NP potential balancing. But it’s not
suitable for application. For simpleness, some important zero- The span of average neutral current is denoted by io' and io'' ,
sequence voltages are selected and compared. The parameter which are the neutral currents when v0  vmin0 and
L and k were defined to represent the serial number of each v0  vmid0 .
switch sequence and the ratio of duration time between the
duplicate voltage vector[21-23]. The zero-sequence voltages io'  vmax 0  iv max  vmid0  ivmid  vmin0  iv min (12)
which can be easily obtained when k  0 and L  0,1, 2, (at
least one phase output voltage is integer level) can be selected io''  vmax 0  iv max  vmid0  ivmid  vmin 0  iv min  2  vmid0  iv min
and intercompared, then the most appropriate one will be (13)
chosen. This process is called searching-optimization strategy.
It can be seen from Fig.3 that, for a three level converter, there (3). vmax0  v0  vmid0 , then vmin  0 , vmid  0 ,
will be five zero-sequence voltages for comparison in the vmax  0 . The average neutral current is:
central meshed region, comparatively, three in the white region
and two in the outside gray region. Obviously, the result is only io  vmax  iv max  vmid  ivmid  vmin  iv min
a comparatively optimal one, but the computation is restricted. (14)
The practicality will be shown in the later section.  vmax 0  iv max  vmid0  ivmid  vmin 0  iv min  2  iv max  v0

It’s necessary to further analyze the relationship between The span of average neutral current is denoted by io'' and io''' ,
the neutral current and zero-sequence voltage. For
where io''' is the neutral current when v0  vmax0 .
va 0  vb 0  vc 0  0 , there must be vmax0  0 , vmin0  0 . Only
io'''  vmax 0  iv max  vmid0  ivmid  vmin 0  iv min (15) 0.5 0.4 0.0

ioL
(4). v0  vmax0 , then the reference voltages are all ioL io'''
ioL

Neutral Current
0.0

Neutral Current

Neutral Current
nonpositive. The average neutral current is:
0.0 -0.4

io  vmax 0  iv max  vmid0  ivmid  vmin 0  iv min  io''' (16) io' ioH -0.4 ioH
ioH
It’s noticed that the possible neutral currents are expressed '' io''
i
from (10) to (16). Actually, for different modulation index and -0.5
o
-0.8 -0.8
-0.4 -0.2 0.0 0.2
-1 0 1 -0.8 -0.4 0.0 0.4
phase angle, the four conditions discussed above mightn’t exist Zero-Sequence Voltage Zero-Sequence Voltage Zero-Sequence Voltage

simultaneously. According to the quantity of the reference (a) case 1 (b) case2 (c) case3
voltages, it’s not difficult to classify the all possible states to Fig 4. Average neutral current (p.u.) caused by zero-sequence voltage.
three cases:
It can be seen from Fig.4 that all available average neutral
Case 1. vmax0  vmin0  1 current values wouldn’t exceed the maximum and minimum
values of the above-mentioned five points. For an expected
In this case, the reference voltages can be all nonpositive neutral current io , it needs to determine which line segment the
and nonnegative with injected zero-sequence voltage. The
curve of neutral current versus zero-sequence voltage is shown neutral current locates in firstly. For example, in case 1,
in Fig. 4 (a). It can be seen that the curve is composed of two because io'  ioH , io'''  ioL , it only needs to identify that which
line segments which are linear with the zero-sequence voltage interval io is in, either between P2 and P3 or between P3 and
and two line segments which are not varied with it. Specially, P4. Once the interval between Px and Py is confirmed, it’s easy
ioL and ioH denote the average neutral currents when to calculate the corresponding zero-sequence voltage using
v0  1  vmin0 and v0  1  vmax0 . Here, the four conditions of interpolation method as follows:
neutral current are all exist in this case, and ioL  io''' , ioH  io' v0Py (io  ioPx )  v0Px (io  ioPy )
v0  (17)
Case 2. vmax0  vmin0  1 and | vmid0 |  max{| vx 0 |}  1 , ioPy  ioPx
where max{| vx 0 |} is the maximum absolute values of the Practically, the difference can be calculated:
reference voltages.
ioPx_dif  io  ioPx , ioPy_dif  io  ioPy (18)
In this case, the reference voltages can’t be all nonpositive
and nonnegative ( m  0 is not considered), but vmid can alter If ioPx_dif  ioPy_dif  0 , then it can be confirmed that io
its polarity. As shown in Fig.4 (b), only two line segments locates in the interval between Px and Py, and the
linear with the zero-sequence voltage remain. io' and io''' can’t corresponding zero-sequence voltage is:
be acquired any more. Average neutral current is confined by
v0py  ioPx_dif  v0px  ioPy_dif
ioL , io'' and ioH . v0  (19)
ioPx_dif  ioPy_dif
Case 3. | vmid0 |  max{| vx 0 |}  1 .
For simply description, we call this strategy precise
In this case, the injected zero-sequence voltage can not alter calculation algorithm. It is noted that, io may locate in two line
the polarity of any phase voltage. Only ioL and ioH can be segments simultaneously in case 1 and case 2. Both results are
obtained, and the curve of average neutral current versus zero- acceptable. Of course, io also may locate in none of the line
sequence voltage is a line segment, shown in Fig.4 (c). segments. At this time, the neutral current which is most closed
According to the aforementioned analysis, there are five to the expected neutral current will be selected. Actually,
significant points formed by the zero-sequence voltage and ioP1_dif , ioP5_dif and ioP3_dif (if P3 is available) are calculated,
corresponding neutral current: and the one that has the minimum absolute value can be
considered as the optimum solution. Fig. 5 is the flow chat of
P1: v0P1  1  vmin0 , ioP1  ioL ;
the proposed algorithm. The exact zero-sequence voltage that
is thought as the most optimal for NP potential balancing can
P2: v0P2  vmax0 , ioP2  io''' ; be gained exactly and conveniently.
P3: v0P3  vmid0 , ioP3  io'' ;
IV. SIMULATION AND EXPERIMENTAL RESULTS
P4: v0P4  vmin0 , ioP4  io' ; In this section, some simulation results will be given to
verify the proposed method. The parameters that used in
P5: v0P5  1  vmax0 , ioP5  ioH . simulation are listed as follows: Vdc  216V ,
I Load (rms )  2.5A , C  740μF , f sw  4kHz , f  20Hz .
START

For va 0 ,vb 0 ,vc 0 obtained from the modulation, defining Calculating the anticipant neutral
vmax0 ,vmid0 ,vmin0 and the corresponding iv max ,ivmid ,iv min , c u r r e n t io a c c o r d i n g t o t h e
calculating the key neutral currents: ioL ,ioH ,ioP3 . departure of the NP potential.

Calculating ioL_dif ,ioH_dif ,ioP3_dif .

vmax0  vmin0  1 ? N | vmid0 |  max{| vx 0 |}  1? N

Y Y

N N N N N
ioL_dif  ioP3_dif  0? ioH_dif  ioP3_dif  0 ? ioL_dif  ioP3_dif  0 ? ioH_dif  ioP3_dif  0 ? ioL_dif  ioH_dif  0 ?

Y Y Y Y Y
v1  v0P2 , i1  ioL_dif v1  v0P3 , i1  ioP3_dif v1  v0P1 , i1  ioL_dif v1  v0P3 , i1  ioP3_dif v1  v0P1 , i1  ioL_dif
v2  v0P3 , i2  ioP3_dif v2  v0P4 , i2  ioH_dif v2  v0P3 , i2  ioP3_dif v2  v0P5 , i2  ioH_dif v2  v0P5 , i2  ioH_dif

v2  i1  v1  i2 Selecting the zero-sequence voltage which is correspond


Calculating the zero-sequence voltage v0  .
i1  i2 to the minimum absolute value of ioL_dif , ioH_dif and ioP3_dif .

END

Fig. 5 Flow chat of the proposed algorithm


110 110 110 110
NP potential

NP potential

NP potential
NP potential

105 105 105 105


0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10
zero-sequence voltage
zero-sequence voltage

zero-sequence voltage

zero-sequence voltage
0.5 vo vo_max vo_min 0.7 0.5 0.7
vo vo_max vo_min vo vo_max vo_min

0.0 0.0 0.0 0.0

-0.5 -0.7 -0.5 -0.7


0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05
time(s) time(s) time(s) time(s)

(a) m=1, =0 (b) m=0.8, =0 (a) m=1, =0 (b) m=0.8, =0
120 120 120 120
NP potential
NP potential

NP potential

NP potential

110 110 110 110

100 100 100 100


0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10
zero-sequence voltage

zero-sequence voltage
zero-sequence voltage

zero-sequence voltage

0.5 0.7 0.5 0.7


vo vo_max vo_min vo vo_max vo_min vo vo_max vo_min vo vo_max vo_min

0.0 0.0 0.0 0.0

-0.5 -0.7 -0.5 -0.7


0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05
time(s) time(s) time(s) time(s)

(c) m=1, =/2 (d) m=0.8, =/2 (c) m=1, =/2 (d) m=0.8, =/2
Fig 6. NP potential and zero-sequence voltage with different modulation Fig 7. NP potential and zero-sequence voltage with different modulation
index and power factor, when the searching-optimization algorithm is used. index and power factor, when precise calculation algorithm is used

The elements that have effect on the NP potential (m=1 and 0.8) are selected to analyze because the fluctuation
fluctuation have been studied, and the main conclusion is: of NP potential is not obvious when modulation index is small.
besides the capacitance of DC capacitor, load current, and the The injected zero-sequence voltages obtained by the
length of the control period, the modulation index and the corresponding methods are also shown in the simulation results.
power factor can also influence the NP potential ripple; the NP The maximum and the minimum zero-sequence voltage (see
potential fluctuation will be severer when the modulation index equation(7)) that can be injected are given simultaneously.
is closer to 1 and the power factor angle is closer to ±π/2[7, Obviously, the NP potential can be maintained steady
24, 25]. Fig.6 and Fig.7 show the NP potential when the whichever algorithm is used, even though the ripple is in
searching-optimization and the precise calculation algorithm evidence when =/2.
are used. Two typical power factors (power factor angle =0 It’s noticed that, when the fluctuation of NP potential is
and /2) are considered. Furthermore, two modulation indexes severe, such as the modulation index is large (m=1) or the
zero-sequence voltage

zero-sequence voltage

zero-sequence voltage
zero-sequence voltage

0.5 0.7 0.5 0.6

0.0 0.0 0.0 0.0

-0.5 -0.7 -0.5 -0.6


0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05
time(s) time(s) time(s) time(s)

(a) R-L load, M=1.0 (b) R-L load, M=0.8 (a) R-L load, M=1.0 (b) R-L load, M=0.8

zero-sequence voltage
zero-sequence voltage
zero-sequence voltage

zero-sequence voltage

0.5 0.5 0.5 0.5

0.0 0.0 0.0 0.0

-0.5 -0.5 -0.5 -0.5


0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05
time(s) time(s) time(s) time(s)

(c) IM load, M=1.0 (d) IM load, M=0.8 (c) IM load, M=1.0 (d) IM load, M=0.8
Fig 8.Experimental results when the searching-optimization algorithm is
Fig 9. Experimental results when the precise calculation algorithm is used.
used.
(The converter operates at 20Hz with R-L load or IM load. For each illustration, respectively, DC-link voltage (blue), NP potential (red), line voltage of
converter (green), line current of the converter (pink) and the injected zero-sequence voltage (black) are shown from the top down.)

power factor is low (=/2), the injected zero-sequence the variation of the DC-link voltage can’t be neglected. Under
voltages calculated by the two algorithms are nearly the same. the experimental condition mentioned above, if the inverter
It can be thought that in this instance, the control margin of NP operates in high power factor, the fluctuation of NP potential is
potential is small, so generally the maximum or the minimum inconspicuous, comparing the fluctuation of DC-link voltage.
zero-sequence voltage will be selected as the injected one. It’s The peak-to-peak value of the NP potential fluctuation is also
evident in Fig.6 (a), (c), (d) and Fig.7 (a), (c), (d). Otherwise, measured when the inverter operates in low power factor.
the precise calculation algorithm can obtain the more precise Slight improvement when the precise calculation algorithm is
result than the searching-optimization one, which can be used can be seen from the corresponding figures. Considering
compared in Fig.6 (b) and Fig.7 (b). Nevertheless, in this the nonideal factor in measurement, the improvement can not
instance, the fluctuation of NP potential is small, and the be thought as the convincing argument.
improvement of the precise calculation algorithm is not
Not only the stable but also the dynamic instances are
significant. However, it can not be denied that the precise
considered. Shown in Fig.10, the NP potential is controlled on
calculation algorithm is the more accurate one.
one third of the DC-link voltage by the presented algorithm.
The presented algorithms are also validated by experiments. Then a command is given to the controller, and the NP
A CM15YE13-12H IPM is used to build the little three-level potential is controlled to one half of the DC-link voltage
NPC inverter. The capacitor in dc-link is 740 μF , and the dc- rapidly, shorter than two operating periods.
link voltage is approximately set to 216V. TMS320F2812 is Although the precise calculation algorithm is much better
used as the main controller. Two operating states are than the searching-optimization one through the theoretical
considered. One is that a three phase load consisting of analysis, the advantage isn’t exhibited evidently in the
inductance (4.2mH) and resistance (31.3  ) is connected to the simulation and experimental results. Besides of the reason
inverter. The power factor is 0.9999 theoretically when the referred to when the simulation results are analyzed, there may
converter operates at 20Hz with the R-L wye-connected load, be some other assignable causes, such as measuring and
representing the condition of high power factor. The other one numerical errors, delay of the digital control, inaccuracy of the
is that the inverter feeds an induction motor with no load. The parameters, etc. It also can be seen that the injected zero-
parameters of the IM are: 2.2kW, 380V, 5A, 50Hz. Because sequence voltages shown in Fig.8 and Fig.9 are not identical
the DC-link voltage of inverter is approximately set to 216V, with the simulation results in Fig.6 and Fig.7. Therefore, the
when the no-load motor operates at 20Hz with rated flux, the two presented algorithms are both good choices, for there may
modulation index is about unit, and the power factor is about not be obvious distinctions in the control performance of NP
0.085, representing the condition of low power factor. potential between them. But the precise calculation algorithm is
Fig.8 and Fig.9 show the experimental results when the the proposed one by the author.
searching-optimization algorithm and the precise calculation Furthermore, the comprehensive analysis of the relationship
algorithm are utilized. Similar with the simulation results, the between the neutral current and zero-sequence voltage can be
NP potential ripple is suppressed satisfactorily except at high used to study the NP-fully controlled region[13]. The concept
modulation index, low power factor. For a passive front is used, has been presented in [7], in which it was also called ripple-



Modulation Index







(a) Searching-optimization algorithm     

Fig. 11 NP-fully controlled region .

(b) Precise calculation algorithm


Fig 10. NP potential is controlled from one third to half of the DC-link
voltage

free area. In order to achieve the so-called fully controlled for


one modulation index and one power factor, the neutral current
must be keep zero. According to the proposed algorithm, the
(a)
acquired region is shown in Fig.11, which is most similar to the 
one shown in [13] but has little difference with the one shown m=1
Maximum NP potential ripple(V)

in [7]. Similarly, the normalized NP potential ripple is shown 


m=0.9
in Fig.12, which is defined as (20) similar to that defined in
m=0.8
[24]. 
m=0.7

u  m=0.6
v  (20)
I rms / fC


V. CONCLUSION 
    
Power factor angle(φ)
A focus in the research of the three-level NPC converter,
problem of NP potential balancing is studied comprehensively (b)
in this paper. For the relationship between the neutral current Fig. 12 Normalized amplitude of the NP potential ripple
and the injected zero-sequence voltage is not linear and
[2] G. Abad, M. A. Rodriguez, and J. Poza, "Three-Level NPC Converter-
continuous, a detailed and deep discussion is necessary. Thanks Based Predictive Direct Power Control of the Doubly Fed Induction
for the achievement that has been acquired and presented in Machine at Low Constant Switching Frequency," IEEE Trans. Ind.
literatures, the relationship is further analyzed in this paper and Electron., vol. 55, pp. 4417-4429, 2008.
made more clearly. Based on that, a searching-optimization [3] M. Malinowski, S. Stynski, W. Kolomyjski, and M. P. Kazmierkowski,
algorithm and a theoretical optimum algorithm employing "Control of Three-Level PWM Converter Applied to Variable-Speed-
interpolation method to precisely calculate the most appropriate Type Turbines," IEEE Trans. Ind. Electron., vol. 56, pp. 69-77, 2009.
zero-sequence voltage to control the NP potential are presented. [4] J. D. Barros and J. F. Silva, "Optimal Predictive Control of Three-Phase
NPC Multilevel Converter for Power Quality Applications," IEEE Trans.
Simulation and experiments are shown to verify the validity Ind. Electron., vol. 55, pp. 3670-3681, 2008.
and the practicability of the proposed algorithms.
[5] J. Rodriguez, S. Bernet, W. Bin, J. O. Pontt, and S. Kouro, "Multilevel
Comparatively, the precise calculation algorithm that is based Voltage-Source-Converter Topologies for Industrial Medium-Voltage
on the thorough and mathematic analysis of the relationship Drives," IEEE Trans. Ind. Electron., vol. 54, pp. 2930-2945, 2007.
between the neutral current and zero-sequence voltage, is more [6] W. Lixiang, W. Yuliang, L. Chongjian, W. Huiqing, L. Shixiang, and L.
superior and valuable, especially in theory. Actually, the two Fahai, "A novel space vector control of three-level PWM converter," in
presented algorithms both show the excellent performance in Proc. IEEE PEDS, 1999, pp. 745-750.
controlling NP potential of three-level NPC converters. [7] N. Celanovic and D. Boroyevich, "A comprehensive study of neutral-
point voltage balancing problem in three-level neutral-point-clamped
voltage source PWM inverters," IEEE Trans.Power Electron, vol. 15, pp.
REFERENCES 242-249, 2000.
[8] K. Yamanaka, A. M. Hava, H. Kirino, Y. Tanaka, N. Koga, and T.
[1] A. Nabae, I. Takahashi, and H. Akagi, "A new neutral-point-clamped Kume, "A novel neutral point potential stabilization technique using the
PWM inverter," IEEE Trans. Ind. Appl., vol. IA-17, pp.518-523, information of output current polarities and voltage vector," in Conf. Rec.
Sept./Oct. 1981. IEEE IAS Annu. Meeting, 2001, pp. 851-858.
[9] S. Busquets-Monge, J. D. Ortega, J. Bordonau, J. A. Beristain, and J. Wang Chenchen (M’09) was born in Anhui
Rocabert, "Closed-Loop Control of a Three-Phase Neutral-Point-
Clamped Inverter Using an Optimized Virtual-Vector-Based Pulsewidth Province, China, in 1981. He received the B.S. and
Modulation," IEEE Trans. Ind. Electron., vol. 55, pp. 2061-2071, 2008. Ph.D. degrees in electrical engineering from
[10] B. Abdul Rahiman, G. Narayanan, and V. T. Ranganathan, "Modified Tsinghua University, Beijing, China, in 2003 and
SVPWM Algorithm for Three Level VSI With Synchronized and 2008 respectively .Now he is a instructor in School
Symmetrical Waveforms," IEEE Trans. Ind. Electron., vol. 54, pp. 486-
494, 2007. of Electrical Engineering, Beijing Jiaotong
[11] S. Ogasawara and H. Akagi, "Analysis of variation of neutral point University, Beijing, China. His research interests are
potential in neutral-point-clamped voltage source PWM inverters," in multilevel converters and motor control.
Conf. Rec. IEEE IAS Annu. Meeting, 1993, pp. 965-970.
[12] C. Newton and M. Sumner, "Neutral point control for multi-level
inverters: theory, design and operational limitations," in Conf. Rec. IEEE
IAS Annu. Meeting, 1997, pp. 1336-1343.
[13] S. Qiang, L. Wenhua, Y. Qingguang, X. Xiaorong, and W. Zhonghong, LI Yongdong was born in Hebei province, China, in
"A neutral-point potential balancing algorithm for three-level NPC 1962. He received the BS. Degree in Electrical
inverters using analytically injected zero-sequence voltage," in Proc.
IEEE APEC, 2003, pp. 228-233. Engineering from Harbin Institute of Technology,
[14] L. Jun, Q. H. Alex, Q. Zhaoming, and Z. Huijie, "A novel carrier-based China, in 1982, and the M.S. and Ph.D degrees in
PWM method for 3-level NPC inverter utilizing control freedom Electrical Engineering from Institut National
degree," in Proc. IEEE PESC, 2007, pp. 1899-1904.
Polytechnique de Toulouse, France, in 1984 and
[15] A. Videt, P. Le Moigne, N. Idir, P. Baudesson, and X. Cimetiere, "A
1987 respectively
New Carrier-Based PWM Providing Common-Mode-Current Reduction
and DC-Bus Balancing for Three-Level Inverters," IEEE Trans. Ind. He was a postdoctoral researcher from 1988 to
Electron., vol. 54, pp. 3001-3011, 2007.
1990 and an associate professor from 1991 to 1996
[16] J. Pou, J. Zaragoza, P. Rodriguez, S. Ceballos, V. M. Sala, R. P. Burgos,
in Department of Electrical Engineering of Tsinghua University. He has been
and D. Boroyevich, "Fast-Processing Modulation Strategy for the
Neutral-Point-Clamped Converter With Total Elimination of Low- a professor since 1996 in the same department of Tsinghua University. He is
Frequency Voltage Oscillations in the Neutral Point," IEEE Trans. Ind. currently responsible of Laborotory of Power Electronics and Motor Control
Electron., vol. 54, pp. 2288-2294, 2007.
of Tsinghua University, Senior Member of China Electrotechnique Society,
[17] J. Zaragoza, J. Pou, S. Ceballos, E. Robles, P. Ibaez, and J. L. Villate,
"A Comprehensive Study of a Hybrid Modulation Technique for the Vice chairman of China Power Electronics Society, Vice chairman of
Neutral-Point-Clamped Converter," IEEE Trans. Ind. Electron., vol. 56, Electrical Automation Committee of China Automation Association.
pp. 294-304, 2009.
His main interests include control theories, sensorless drives and
[18] J. Zaragoza, J. Pou, S. Ceballos, E. Robles, C. Jaen, and M. Corbalan,
"Voltage-Balance Compensator for a Carrier-Based Modulation in the applications of AC motors; high-voltage, high-power inverter for motor drives
Neutral-Point-Clamped Converter," IEEE Trans. Ind. Electron., vol. 56, and active power filter application.
pp. 305-314, 2009.
[19] V. Blasko, "Analysis of a hybrid PWM based on modified space-vector
and triangle-comparison methods," IEEE Trans. Ind. Appl., vol. 33, pp.
756-764, 1997.
[20] W. Fei, "Sine-triangle versus space-vector modulation for three-level
PWM voltage-source inverters," IEEE Trans. Ind. Appl., vol. 38, pp.
500-506, 2002.
[21] H. Xuan, L. Yongdong, and L. Yongheng, "A novel general space vector
modulation algorithm for multilevel inverter based on imaginary
coordination," in Proc. IEEE PEDS, 2003, pp. 392-396.
[22] L. Yongdong, G. Yue, and H. Xuan, "A general SVM algorithm for
multilevel converters considering zero-sequence component control," in
Proc. IEEE IECON, 2005, pp. 508-513.
[23] C. Wang, Y. Li, and X. Xiao, "A Unified SVM Algorithm for Multilevel
Converter and Analysis of Zero Sequence Voltage Components," in
Proc. IEEE IECON, 2006, pp. 2020-2024.
[24] J. Pou, R. Pindado, D. Boroyevich, and P. Rodriguez, "Evaluation of the
low-frequency neutral-point voltage oscillations in the three-level
inverter," IEEE Trans. Ind. Electron., vol. 52, pp. 1582-1588, 2005.
[25] C. Wang, "Research on the topology, PWM algorithm and balance
control of neutral point voltage in multilevel converters." Dissertation
for degree of doctor, Tsinghua University, 2008.