Active Power Filtering by a Flying-Capacitor

Multilevel Inverter with Capacitor Voltage Balance

Junfei Hu, Li Zhang, S. J. Watkins
School of Electronics and Electrical Engineering, University of Leeds, UK
[email protected]

Abstract- A new control strategy is developed for a four-level  Incorporating the rate of change of the reference current
flying capacitor multilevel inverter to function as an active power in the control algorithm to improve the accuracy in
filter (APF). This employs space vector modulation (SVM)
control at transient current changes.
scheme. However inverter switching vectors and sequences are
determined according to the swings of capacitor voltage values at The proposed FCMI-APF has been simulated to eliminate
each level. A modified predictive current control law is developed the load current harmonics in a small power system supplying
This takes into account the derivative of the reference current in a nonlinear load. The universal program system EMTP-ATP
order to improve the performance of the conditioner when (Electromagnetic Transient Program–Alternative Transient
reference current changes quickly. Simulation study has shown
Program.) has been used to conduct this study. The proposed
that the proposed scheme leads to high performance harmonic
current compensation for both steady and transient state control strategy has led to good performance harmonic
operations and capacitor voltages at all levels are well balanced. elimination, particularly for transient operations.
The paper describes the principles of the APF control
I. INTRODUCTION scheme. The specific SVM scheme for a four-level flying-
capacitor multilevel inverter is presented. Simulation result of
Active Power Filters (APF) has been used to eliminate
the FCMI-APF controlling a power system will be presented
harmonic currents control reactive power flow in power
and the features of the method in comparison to the other
systems. They are going to be indispensable for maintaining a
approaches will be highlighted.
high quality power supply and will play an important role in
the future due to the increasing integration of electric power II. THE SYSTEM
generators powered by renewable energy sources within the
The power system simulated is shown in Figure 1. The three-
power distribution infrastructure.
phase power source is separated from the load by the series
Conventional two-level inverter topology cannot be applied
impedance LS-RS. The nonlinear load is represented by a three-
directly to high voltage power distribution systems which have
phase thirsted controlled rectifier supplying an R-L load, and a
voltage levels of tens of kilovolts because of limited device V/I
three-phase load filter characterized as LL-RL is used. The APF
ratings and high dV/dt. A viable approach to high power
is a four-level (including 0 voltage level) flying-capacitor
conversion is by constructing multilevel inverters, in particular,
multilevel inverter as shown in Figure 1. It is shunt connected
the flying capacitor multilevel inverter (FCMI) proposed by
to the power line common coupling point (CCP) on the load
Meynard, et al [1,2, 3]. The FCMI has been used as a shunt
side through a filter featured as Lf-Rf. This is necessary for
power line conditioner and also as a unified power flow
eliminating current ripples due to switching.
controller (UPFC) [3, 4]. The main challenge in using an FCMI
as an APF lies in obtaining fast and accurate performance in
harmonic compensation whilst maintaining the voltage across
the floating or cell-capacitors at the appropriate constant levels.
Large capacitor voltage swings may cause voltage stress across
devices, consequently poor harmonic control performance and
even inverter failure. Hitherto few published works have
addressed this issue.
This paper investigates a new control scheme for a four-level
FCMI functioning as a shunt active power filter. It employs a
predictive current controller, and uses Space Vector
Modulation (SVM) method to determine the inverter switching
vectors and their corresponding duty cycles [5], the proposed
control strategy has two novel features:
 employing capacitor voltage feedback signals to
determine the inverter switching vectors and sequence,
and
Figure 1. Schematic of the power system and a FCMI-based shunt APF.

978-1-4244-1666-0/08/$25.00 '2008 IEEE 2348

 III. THE PREDICTIVE CONTROL PRINCIPLE Using a low-pass filter shown in Figure 2 we can extract the
As is well known the APF control scheme consists of two fundamental current. Its equivalent AC quantity iLaf, iLbf, iLcf can
processes. The first extracts the unwanted harmonic be derived by applying inverse Park’s transform. Thus the load
components from the measured load current. These are taken harmonic current components can be calculated as
as the reference values for the second part which implements a iLah  iLa  iLaf 
i   i  i  (3)
feedback control scheme for canceling them from the power  Lbh   Lb Lbf 
line current. iLch   iLc  iLcf 
There are a number of methods for the first task; one of them
is the instantaneous reactive power theory [7, 8]. However for where iLa , iLb and iLc are the three phase load current and iLah ,
simplicity the synchronous reference frame algorithm (SRF) iLbh and, iLch are the harmonic components.
[9, 10] is applied here. This assumes that the three phase These extracted harmonic currents are applied as the
voltages of the simulated power system are balanced and reference currents for the APF control loop as shown in Figure
largely noise free hence they are represented as a space vector 2(b). The aim of APF controller is to eliminate the harmonic
rotating around a stationary reference frame at the mains current components from the load current. A predictive current
frequency. If the measured three-phase current vector is control algorithm combining with the SVM scheme for FCMI
projected to a α-β plane, phase locked to the rotating voltage is employed. The principle of predictive control scheme is well
vector, according to Park’s transformation it can be expressed known. Having a model of the system to be controlled it
as estimates the appropriate control action to enable controlled
variable to follow the desired value.
 i p   sin  t  cos  t   i  (1) In this application the model of the system is the R-L
i     
 q    cos  t  sin  t  i   impedance between the inverter output and power line
common coupling point. The input and output variables are the
where iα , i are the two orthogonal

current components of the inverter ac terminal voltage and current vectors, respectively.
three phase current vector I in SRF, ip and iq are defined as Thus the equation expressing the relationship between the
instantaneous active and reactive currents respectively and  is input and output variables is given as
the fundamental angular frequency of the AC system.    
e  V f  L f  ddtiL  R f  i C (4)
Both ip and iq are comprised of two components, the
fundamental frequency component transformed to a DC where Rf and Lf are the line/filter
 impedance between inverter
quantity and the harmonic component as expressed below ac terminal and CCP, iC is the inverter current vector
~ (2) equivalent
 to the compensating current supplied by the APF,
i  i  i  
p
i   
p p
~ e and Vf are the voltage vectors at CCP and the inverter ac
 q   iq  iq 
terminal respectively. d i dt is the rate of change of current
between two samples. During a short and finite switching
period Ts, this can be approximated by ratio of difference terms
as
  
 iC i C ( t K  T s )  iC ( t K )
Ts  Ts (5)

Substituting iC ( t k  T s ) with the estimated reference th
current and iC ( t k ) with the measured  current value at k time
interval, the ac side voltage vector Vf at next sample can then
be calculated as:
(a)Harmonic current extraction   L  L 
V f (t K  TS )  e (t k )  Tsf iC* (t k )  ( R f  Tsf )  iC (t k )
(6)
It is worth noting that the above algorithm does not work
well under operating conditions where reference current
changes rapidly. This is due to the natural delays in sampling
steps caused by both the reference current and inverter
reference voltage calculation processes.
This effect is revealed in this particular application when the
firing angles of the three-phase rectifier become greater than
30°; the source current waveform contains undesirable sharp
(b)Predictive current control spikes as shown in Figure 5. To overcome this defect the new
Figure 2. Control system block diagram
modified algorithm incorporates the rate of change of the

2349
reference current at the current and last samples into Equ(6), negative DC rail, VaN, can be synthesized to the following
subsequently we have voltage level:
   
  L      
L
Vf  e(tk )  Tsf iC* (tk )  (Rf  Tsf )  iC (tk )  n  iC* (tk )  iC (tk )  m iC* (tk )  iC* (tk1 ) 1) For voltage level 0, VaN=0
2) For voltage level 1, VaN=Vdc/3.
(7) 3) For voltage level 2, VaN=2Vdc/3.
where m and n are the weighting factors for newly added 4) For voltage level 3, VaN=Vdc.
terms. Similar to adding derivative terms in a P+I controller, The triangle is formed by state vectors Va, ,Vb, and Vz . The
these terms enable the controller responding promptly to active time for the state vectors can be given by the following
reference current variations. However they also make the formulae,
controller sensitive to measurement noise, which may lead the    
system unstable. Thus it is important to tune m and n factors Ta  V a  Tb  Vb  T z  V z  Ts  V (8)
carefully to achieve the desired transient performance. T S  Ta  Tb  T z (9)
Normally, m and n are set with 10 and 5 respectively. where Ts is the carrier period, Ta, is the time interval for Va , Tb
and Tz are the time intervals for Vb, and Vz, respectively, V is
the target output space vector. The duty time equations for
sectors 1 to 6 are listed in Table II, where k is equal to 3 and M
is the modulation index, defined as the ratio between the
magnitudes of the modulating waveform and the carrier signal.
TABLE I
SPACE VECTORS FOR A FOUR-LEVEL INVERTER

Level 1 Level 2
V S V S V S
0; 0; 0
1; 1; 1 2; 0; 0 0; 2; 2
V0 V200 3; 1; 1
V022
2; 2; 2 1; 3; 3
3; 3; 3

1; 0; 0
2; 1; 0 0; 1; 2
V100 2; 1; 1 V210 V012
3; 2; 1 1; 2; 3
3; 2; 2

Figure 3. Switching vectors in a multilayer hexagon 1; 1; 0

2; 2; 0 0; 0; 2
V110 2; 2; 1 V220 V002
3; 3; 1 1; 1; 3
3; 3; 2
IV. SPACE VECTORS MODULATION SCHEME
0; 1; 0
Once obtaining inverter voltage vector the SVM scheme is 1; 2; 0 1; 0; 2
V010 1; 2; 1 V120 V102
2; 3; 1 2; 1; 3
applied to determine the switching vectors and calculate their 2; 3; 2
corresponding durations. The method of vector selection is 0; 1; 1
0; 2; 0 2; 0; 2
described below taking a four-level three-phase FCMI as the V011 1; 2; 2 V020 1; 3; 1
V202
3; 1; 3
example. 2; 3; 3

In this case there are, in total, 40 switching vectors, 36 active 0; 0; 1

0; 2; 1 2; 0; 1
and 4 zero ones. The spread of these vectors within a bounded V001 1; 1; 2 V021
1; 3; 2
V201
3; 1; 2
2; 2; 3
three-layer hexagon is shown in Figure 3. Selection of the
switching vectors at every sample interval is based on the 1; 0; 1

location of Vref in the hexagon, which is coded according to the V101 2; 1; 2

3; 2; 3
54 small triangles (see Figure 3), but also importantly on the
measured capacitor voltage levels. Each of these triangles has Level 3
three adjacent active switching vectors. Those of them V S V S V S
connecting the middle and low voltage levels involve either a V300 3;0;0 V030 0;3;0 V003 0;0;3
capacitor being charged by the power source or the other being V310 3;1;0 V031 0;3;1 V103 1;0;3
discharged through the load. A look-up table is set up to store V320 3;2;0 V032 0;3;2 V203 2;0;3
V330 3;3;0 V033 0;3;3 V303 3;0;3
the charging and discharging features of all the switching V230 2;3;0 V023 0;2;3 V302 3;0;2
vectors. V130 1;3;0 V013 0;1;3 V301 3;0;1
The vectors are arranged according to their layers from 1 to
3. A vector table is established with the phase leg output Using the capacitor voltage feedback signal we examine the
voltage levels as shown in Table I. For example, the voltage of extent of relevant capacitor voltage deviations from their
the four-level inverter phase leg a output with respect to the respective desired levels. Subsequently we determine the

2350
switching vectors and their sequence of implementation in the the switch combination selected will cause it to discharge
inverter appropriate for minimizing capacitor voltage swings. through the load.
The above operation is included in block CV in Figure 2. V. SIMULATION RESULTS
Calculation of the duty cycles for the selected vectors is then
followed as indicated in block CD/CS in the figure 2. The above described control method has been applied to the
TABLE II
simulated power system shown in Figure 1. The voltage rating
is set to 600V (peak-peak), 50 Hz and current rating 150A. The
FORMULAE FOR DUTY CYCLE CALCULATION
power line impedance values are defined as Rs = 0.1Ω and Ls =
Sector, Phase Vector Sector Duty Equation
0.0085mH. The load resistance R is 2 Ω. The load input filter
n Angle, θ parameters are Rr = 0.1 Ω and Lr = 0.3mH. The resistance and
1 0≤θ< Vka0, Ta  2
3
k  M  Ts  cos  6   (a  k  1)Ts inductance of the APF filter are 0.1 Ω and 0.8mH, respectively.
/3 Vkb0, The results of simulation are shown in Figures 4-6. In Figure 5,
Vza0 Tb  2
3
k  M  T s  sin   a  T s
the modified algorithm is compared with the method when
Vzb0, Ta  3
k  M  Ts  sin   3   (k  a  1)Ts
Vkb0,
2
derivative terms are not used. It is shown clearly that the latter
Vza0 Tb  3
k  M  Ts  cos(  6 )  (k  1)  Ts
2
leads to spikes on the peak points of current waveform. In
2 /3 ≤ θ Vak0,
Ta  3
k  M Ts  sin(  23 )  (k  a) Ts contrast, when using the proposed method, these spikes are
Vbk0, 2
< 2/3
Vaz0 Tb  2
3
k  M  Ts  cos   6   (1  a )Ts significantly attenuated giving a superior current waveform.
Vak1, Ta  3
k  M  Ts  cos(  2 )  ( k  1)  Ts The waveforms in Figures 6(a)-(f) are for transient current
2
Vak0, responses with α changing from 0˚ to 30˚and from 30˚ to 60˚.
Vaz0 Tb  2
3
k  M Ts  sin  23   a Ts Figures 6(a) and (d) show load current, (b) and (e) give APF
3 2/3 ≤ θ V0ka, Ta  2
3
k  M  T s  cos   2   ( a  k  1)Ts compensating currents respectively for both α values, and (c)
< V0kb,
V0za Tb  2
3
k  M  T s  sin(   2
3
)  a  Ts and (f) are the power line current waveforms. Clearly fast
harmonic compensation has been achieved in both cases using
V0zb,
Ta  3
k  M Ts  sin    (k  a 1)Ts
V0kb, 2
the proposed control scheme. The performance of the proposed
V0za Tb  3
k  M  T s  cos(   5
)  ( k  1)  T s
2 6
APF is satisfactory while the flying-capacitor voltages at both
4 ≤θ< V0ak, Ta  3
k  M  Ts  sin(   )  (k  a)  Ts
4/3 V0bk, 2 levels 1 and 2 are well balanced as shown in Figure 4.
V0az Tb  2
3
k  M  T s  cos   5
6   (1  a )Ts ;
V1ak, Ta  3
k  M  Ts  cos(  7
)  ( k  1)  Ts
2 6
V0ak,
Tb  3
k  M  T s  sin   4
 a Ts
V0az 2 3

5 4/3 ≤ θ Va0k, Ta  2
3
k  M  Ts  cos   7
6
  ( a  k  1)Ts
< 5/3 Vb0k, Tb  3
k  M  T s  sin(   4
)  a Ts
Va0z 2 3

Vb0z, Ta  2
3
k  M  T s  sin   5
3
  ( k  a  1) T s
Vb0k,
Tb  3
k  M  Ts  cos(   3
)  ( k  1)  Ts
Va0z 2 2

6 5/3 ≤ θ Vk0a,
< 2 Vk0b, Ta  23 k MTs sin(  53 ) (k a)Ts
Vz0a Tb  2
3
k  M  T s  cos   3
2  (1  a ) T s
Vk1a, Ta  2
3
k  M  T s  cos(   7
6 )  ( k  1)  T s Figure 4 Balanced flying capacitor voltages
Vk0a,
Tb  3
k  M  T s  sin   4
  a  Ts
Vz0a 2 3
60

[A]

40

IV CAPACITOR VOLTAGE BALANCING CONTROL 20

Since there are two floating capacitors in each inverter limb, 0

the voltage balance of these capacitors (six of them) is a key

-20

issue for the system to operate as desired. In this work, a

-40

-60

voltage feedback scheme is applied. This measures the 0.04 0.05

(f ile c f 10kl4.pl4; x -v ar t) c :UA
0.06
-INSTRA
0 .07 0.08 0.09 [s ] 0.10

capacitor voltages at every sample interval and compared with (a)Power line current with modified algorithm
70 .0

their desired values. The errors are applied to determine the [A]
52 .5

appropriate vector and switching combination at the 35 .0

corresponding sample interval. Especially for voltage level 1 17 .5

0 .0

and 2, there are different switching combination to charge and -17 .5

discharge the capacitor. For example if the voltage of C1 is -35 .0

lower than 1/3 VDC by a predefined error band a switch -52 .5

combination will be chosen such that it leads to C1 being -70 .0

0.04 0.05
(f ile c f 10kl4d.pl4; x -v ar t) c :UA
0.06
-INSTRA
0.07 0.08 0.09 [s ] 0.10

charged, while the inverter output voltage follows the desired (b)Power line current with conventional algorithm
value. On the other hand if voltage of C1 is higher than desired, Figure 5: Simulation results

2351
 120
[A]
150
80
[A]

100
40

50 0

0 -40

-50 -80

-100 -120
1.98 2.00 2.02 2.04 2.06 2.08 2.10 [s ] 2.12
(f ile c f 10kl4ln.pl4; x -v ar t) c :UC -INSTRC
-150
1.98 2.00 2.02 2.04 2.06 2.08 2.10 [s ] 2.12 (f)
(f ile cf 10kl4ln.pl4; x-var t) c:INSTRA -X0267A
Figure 6 Transient response for α from 0˚ to 30˚; ((a) Load current; (b) APF
(a) current; (c)Compensated source current ;
150
[A]
Transient response for α from 30˚ to 60˚ (d)Load current; (e) APF current;
100 (f)Compensated source current

50 VI. CONCLUSIONS
0
The paper presented an FCMI-APF for harmonic
-50 compensation. The predictive control scheme has incorporated
-100
derivative terms and SVM method for inverter switching
control. Capacitor voltage values are feedback to the control
-150
1.98 2.00
(f ile cf 10kl4ln.pl4; x-var t)
2.02
c:X0013A -X0009A
2.04 2.06 2.08 2.10 [s ] 2.12 loop to assist switching vector selection at each sample.
(b) Simulation result has shown that the proposed method is
150
[A]
capable to give desired performance for both steady and
100 transient operations.
50 REFERENCES
0 [1] Meynard, T.A. and Foch, H., "Dipositif électronique de conversion
d'énergie électrique", French Patent No. 2679715B1, Filed 25th July 1991,
-50
Issued 29th January 1993. (WO93/02501A1, “Electronic device for
-100
converting electrical energy”).
[2] Meynard, T.A. and Foch, H., Multi-level choppers for high voltage
-150 applications, EPE Journal, Vol. 2, No. 1, March 1992, pp. 45 – 50.
1.98 2.00 2.02 2.04 2.06 2.08 2.10 2.12
(f ile cf 10kl4ln.pl4; x-var t) c:UA -INSTRA
[s ]
[3] Liang, Y.; Nwankpa, C.O., A power-line conditioner based on flying-
(c) capacitor multilevel voltage-source converter with phase-shift SPWM,
150 Industry Applications, IEEE Transactions on Vol. 36, No. 4, Jul/Aug
[A] 2000, pp. 965-971
100
[4] L. Xu and V.G. Agelidis, Flying capacitor multilevel PWM converter
based UPFC, IEE Proc. Electr. Power Appl., Vol 149, Jul 2002, No. 4, pp.
50
304-310
0
[5] L. Xiao, L.Zhang, “A Space Vector-Based Deadbeat Controller for Shunt
Active power Filters”; EPE’97, pp 4.864-4.869
-50 [6] Watkins, S.J.; Zhang, L.; Multilevel space vector PWM control schemes
for a flying-capacitor inverter; PEMD 2004, Vol. 1, pp:12 - 17
-100 [7] H. Akagi, Y. Kanazawa, and A. Nabae, “Generalized theory of the
instantaneous reactive power in three-phase circuits,” in Proc. I983 Int.
-150
1.98 2.00 2.02 2.04 2.06 2.08 2.10 [s ] 2.12 Power Electronics Conf, Tokyo, Japan, 1983, pp. 1375-1386.
(f ile cf 10kl4ln.pl4; x-var t) c:INSTRA -X0271A
[8] Akagi, H.; “Active Harmonic Filters”; Proceedings of the IEEE Volume
(d) 93, Issue 12, Dec. 2005 Page(s):2128 – 2141
120

[A]
[9] S. Bhattacharya, M. Divan, and B. Benerjee, Synchronous Reference
80 Frame Harmonic Isolator Using Series Active Filter, EPE’91, Vol. 3, pp.
30-35
40 [10] Bhattacharya, S., Divan, D., Synchronous frame based controller
implementation for a hybrid series active filter system, IEEE/IAS Annual
0
Meeting 1995, pp2531-2540
-40

-80

-120
1.98 2.00 2.02 2.04 2.06 2.08 2.10 [s ] 2.12
(f ile cf 10kl4ln.pl4; x-var t) c:X0013A -X0009A

(e)

2352