Nonlinear control technique to enhance dynamic

performance of a shunt active power filter

N. Mendalek, K. AI-Haddad, F. Fnaiech and L.A. Dessaint

Abstract: The modelling of, and a nonlinear control strategy for, a three-phase voltage source
shunt active power filter is presented. The dynamic model is at first elaborated in the system ’abc’
and then transformed into the synchronous orthogonal ‘dq’ frame. The ‘dq’ frame model is divided
into two separate loops, namely the two current dynamics inner loop and the D C voltage dynamics
outer loop. The exact feedback linearisation theory is applied in the design of the controller. Then,
the pole placement strategy is used to synthesise the closed loop error dynamics of current tracking
and DC bus voltage regulation. The adopted control strategy allows the decoupling of the currents
and improves their trackmg behaviour and enhances DC voltage regulation. Simulation results
validate the expected active filter performance in both steady state and transient operation, and also
show that the controller is capable of compensating under severe load current imbalances.

1 Introduction flux based control [5], and nonlinear optimal predictive

control [7]. These methods allow good current tracking
In recent years, the widespread application of nonlinear performance: nevertheless, the outer loop controller is
loads has increased the harmonic-related problems in utility based on the system linearisation around the operating
and industrial power systems. Consequently, active power point with a Proportional-Integral (PI) compensator. In [6],
filters have been investigated and were expected to be a a nonlinear control strategy is employed in both loops.
viable solution [l, 21. For instance, shunt active filters are However, the decoupling of the inner loop is based on an
known to be an appropriate means of eliminating harmonic intuitive selection of the auxiliary inputs. In contrast, in
contamination created by current type nonlinear loads. this paper, we use the theory of the exact linearisation
Shunt passive filters have been traditionally used to absorb technique to decouple the inner loop variables. Thus, we
t h s type of harmonics. They have, however, major will consider a new nonlinear control strategy for the shunt
drawbacks, including the dependency of their compensation active filter to enhance the dynamic response when
characteristics on the utility grid impedance, and their compensating for both harmonic pollution and reactive
susceptibility to undesirable resonance conditions with grid power. Ths method represents another attempt to solve
and load impedances caused by the wide spectrum of the nonlinear control problem by applying an exact
current harmonics generated by the load. linearisation control to both inner and outer loops.
The nonlinear loads such as phase-controlled rectifiers Consequently, the active filter tracks its current references
draw distorted currents which are characterised by sudden exactly and thus behaves as a quasi-ideal current source
slope variations. Therefore, the active filter and its current connected in parallel with the load. It generates the volt-
control should be effective in tracking the dynamic current ampere reactive and the harmonic currents required by the
references, whch makes the power stage highly sensitive load, thereby achieving sinusoidal supply currents locked
and the design of the control particularly critical. The usual with the phase voltages under either dynamic or steady state
approach employed to control the active filter consists in modes of operation.
decoupling the system into two loops [3-71. Thus, the
active filter model is divided into fast inner current loop 2 Active filter model
and slow outer DC voltage loop. An effective control
strategy is then applied to the current loop with the aim 2. I Modelling in the ‘abc‘ reference frame
of achieving fast and stable tracking performance, and a Fig. 1 shows a typical shunt active filter connected in
linear control is applied to the voltage loop to regulate parallel with nonlinear loads. The active filter consists of
its setting point. The control strategies utilised in the a standard six-switch three-phase voltage source inverter
inner loop include sliding mode control in the system ‘abc’ (VSI), input AC inductors (Lc,Rc)and a DC bus capacitor
frame [3] or in the orthogonal synchronous ‘dq’ frame [4], C. The three-phase AC mains system with the line
inductance L, is feeding a three-phase thyristor bridge
rectifier with inductive load. The provision of operating
0IEE, 2003
a single-phase thyristor rectifier load is made with the
IEE Pruceednqs online no. 20030488
switch SW in order to realise an unbalanced nonlinear
doi:10.1049/ip-epa:20030488
load.
Publication date: 21st May 2003. Paper first received 2nd October 2002 and in
revised form 14th February 2003 A dynamic analytical model of the active filter may be
N.Mendalek, K. AI-Haddad and L. Dessaint are with the Dipartement de developed in its original three-phase ‘abc’ frame. In
Genie Electrique, Ecole de Technologie SupCrieure, I 100, rue Notre-Dame applying the general Kirchhoff s voltage and current laws
Ouest, MontrCal, QuCbec, H3C I K3, Canada at the connection point of the active filter, the following
F. Fnaiech is with the ESSIT, 5 AV.Taha Hussein, Tunis 1008, Tunisia three dfferential equations in the stationary ‘abc’ frame
IEE Proc-Electr. Power Appl, Vol. 150, No. 4, July 2003 373
 and it can be verified that:

in= 1 m= I

This allows for obtaining:

+
and, using il + i2 i3= 0, the following differential equation
on the DC side is obtained:

As a result, the complete model of the active filter in the

Fig. 1 Shunt active power filter connected to the electrical system ‘abc’ referential is obtained by employing (4) for phases ‘1’
and ‘2’, and (8):
may be written: L ,4 -
dt - -&il - dnludc VI +

By summing the three equations in (l), and tahng into

account the absence of the zero-sequence in the three wire
{ L C&d t- -- R i
C, du,k -
- (2dnl
C 2 -d n2dc+U2

+ dn2)i1 + ( d n l + 2&2)i2
The state variables of this ‘abc’ frame representation have
(9)

systern currents, and assuming that the AC supply voltages sinusoidal steady state waveforms at the supply frequency
are balanced, one obtains: o.Furthermore, the interaction between the three phases
represents another drawback of the ‘abc’ model. Therefore,
l 3 in order to facilitate control, the model can be transformed
= -- UmM to the synchronous orthogonal frame rotating at the supply
- m=l
?
frequency. With this time-varying transformation, the
The switching function ck of the converter’s kth leg (for positive-sequence components become constant, and the
k = 1, 2, 3) is defined as: effect of the interaction between the phases is avoided at
the switching state decision level.
1, if s k is On and SL is Off
ck={ 0, if s k is Off and S i is On (3)
2.2 The model transformed into the ’d9’
Hence, by writing ukM= ckude, the following relation is reference frame
obtained: The model transformed into the synchronous ‘dq’ frame is
given by:

I +
L ,4-
dt - -&id +L,0iq - V&dnd ud

The relation (4) represents the dynamic equation of the L ,4

dt -
- -Reiq - Lcoid - ukdnq + uq (10)
filter’s model in phase k. Furthermore, we may define a
function dnk that we call switchng state function whch is
C d;”L - dndid + dnqiq
given by the following: The model represented by (10) is nonlinear because of the
existence of multiplication terms between the state variables
{id, iq,ude} and the inputs {dnd,d,zy}. However, t h s model is
time invariant during a given switching state.
Furthermore, the principle of operation of the active filter
Equation (5) shows that the value of d& depends on the
switching state n and on the phase k. In other words, dnk requires that the three state variables have to be controlled
depends simultaneously on the switching functions of the independently. Therefore, each of the currents id and iq has
to follow a varying reference current extracted from the
three legs of the active filter. Ths shows the interaction
nonlinear load currents. In addition, the DC voltage level
between the three phases. Moreover, based on (5) and
vd, has to be regulated at a set point in order to maintain the
for the eight permissible switching states of the active
compensation performance of the active filter during
filter (n = 0, 1,. ..,7), the conversion of [ek] to [dnk]
is given
dynamic variations. However, the model has only two
by:
independent inputs. Hence, the exact feedback linearisation

[bi]
-1 -1 is not applicable, since the corresponding decoupling matrix

[ti]=: [
4x3 2 1 -12 -11
-1
It is worthwhile to note that the rank of the transformation
(6) is not square. One way to overcome this difficulty is to
consider that the model (10) may be divided into two
separate loops, namely the inner fast current loop, and the
outer slow DC voltage loop. The interaction between the
matrix in (6) is two, and that the system [dnk]has no zero- two loops can be avoided by adequately separating their
sequence component. respective dynamics.
On the other hand, on the DC side the following
equation holds: 3 Control strategy

Fig. 2a illustrates the control scheme of the AF. Since there

is no neutral line, only two independent currents require to
374 IEE Proc.-Electr. Power Appl.. Vol. 150,No. 4, July 2003
 ~~

Current controller 1

"wz4
I I
I

k
d p
d
i'd Id
> Tracking ud > Control I' 'nd> Nonlinear
controller law system
~

I
> Eq. (14) uq > Eq. (12) ~ '"4) Eq. (10) 'dc
I
II .~ - ~___.._
T ! --L

Fig. 2
U Detailed block diagram illustrating the shunt active filter with feedback linearizing control
h Bloc diagram of the controlled system

be sensed in the three-phase system. The AF currents il and the following form:
i2 of phases '1' and '2' respectively, are sensed and
transformed to the synchronous reference frame via the
two transformations 1-2/a-D and Cr-pld-q to obtain their
corresponding currents id and iy.These synchronous frame
currents are used in the feedback control of the inner loop.
In the following, the control design of the inner and outer yl = h l ( x ) = i d
loops is presented.
y2 =h2(x) = i,
3.1 Inner loop
In order to design the current loop, the first two equations
in the model are considered. With appropriate input
transformation, the coupled dynamics of the currents Note that system (1 1) is a multiple-input multiple-output
tracking problem is transformed into decoupled dynamics. (MIMO) nonlinear subsystem. In the light of the theory of
The first two equations of the model (10) are rewritten in non-linear control this system has relative degrees
IEE Proc -Electr. Power Appl., Vol. 150, No. 4, July 2003 375
y1 = 1-2 = 1, and the following definitions hold: Afterwards, the required active filter voltage vector that
realises d,, and d , ~is selected from the eight possible
switchmg states with the help of a switching table.

3.2 Outer loop

The instantaneous active power dissipated by the
active filter, and the instantaneous reactive power ex-
changed between the active filter and the supply mains
Clearly A(x) is not singular. Thus, we can deduce the are given by:
input signals to be modulated by the space
p = Udid and q = -udi,, (because U , = 0)
modulator by:
These expressions show that in order to maintain the DC
voltage level at a desired value, the losses through the active
filter components can be compensated by acting on id.
Therefore, an auxiliary input of the output feedback loop is
This gives: defined as:
du;
i, = C-=
dt
C-+k-
dt
+ J'
k2 &dt (18)

Applying these nonlinear feedback controllers, the non- where i& = u:e - U& is the DC voltage error. Similar to the
linear subsystem (11) becomes a simple set of linear inner current loops, the corresponding closed loop dynamic
decoupled differential equations given by: voltage error equation is given by:

];[
did

= [E]
We are able then to design a control law for the
and the proportional ( k , ) and integral (k2) gains are then
computed from the following relations:
simple linear decoupled MIMO system (13). The currents kl = 2to,,C and k2 = w:,C (20)
id and i, can thus be controlled independently by where the outer loop natural frequency w,,, should be
acting upon the inputs ud and U,, respectively. A funda- chosen to be as small as possible (around the
mental issue in deriving the tracking controller is the supply fundamental frequency). Using (1 8) and the third
addition of an integrator in order to attenuate the equation in (lo), the outer loop control effort can be
tracking steady-state error. Hence the tracking expressed as:
controllers are constructed by using the following
expressions:

Moreover, assuming the inner current loops to be ideal and

considering that the active filter operates under normal
conditions, the following properties hold:
where id = :i - id and i, = ii - i, are the current dnqukM U , andddud M Ud (22)
errors, and {i;, ii} denote the references of {id, i,}, On the other hand, the phase voltages at the active filter
respectively. connection point are assumed to be:
The closed loop current error dynamics are given by:
U1 = v cos(ot)
~2 = f' cos(wt - 2 ~ / 3 ) (23)
U 3 = f' cos(0t - 4n/3)

In order to achieve asymptotic tracking, the error Their transformation to the synchronous reference frame
characteristic equations are equated to the usual standard gives:
second order equation, whch is s2 2t0,is w:i = 0, + +
where wni is the natural frequency and ( is the damping
factor. The proportional (kp)and integral (ki)gains are then
obtained as follows:
k, = 250,~ and ki = CO:, (16)
For the optimal value of the damping factor ( = d / 2 , the which yields:
inner loop natural frequency wni may be chosen to be as dqu& M U , = 0 and d,,du& = Ud = m V (25)
high as possible, but is bounded by the switchmg frequency
value. Thus, the contribution of this outer loop is given by the
Several modulation techniques can be applied to realise following:
d,d and d,,<,. We have considered for t h s application the
instantaneous space-vector strategy. The inputs dndand dny
are transformed to the up coordinates using the following
transformation: The reference current in (26) is added to the harmonic

I;:[ =
cos cot
[sin ot
- sin w t ]
cos ot
[21 (17)
reference current of the d-axis loop. The reference current
ii0is a DC component and it forces the active filter to draw
a current at the fundamental frequency in order to
376 IEE Proc.-Electr. Power Appl., Vol. 150. No. 4, July 2003
compensate for the converter switchng and conduction 4.1 Low order current harmonic compensation
losses. Moreover, by appropriately designing the DC In steady state operation, the simulation results show a total
voltage loop to be much slower than the current loop, harmonic distortion (THD) of 4.31 Yn in the supply currents
there will be no interaction between them. In fact, in this when the THD of the load currents is 31.5%. The harmonic
application the corresponding inner and outer loops spectrum of iL and is currents are depicted in Fig. 3.
frequencies are chosen to be 3 kHz and 40 Hz, respectively.
Fig. 2b shows the bloc diagram of the whole controller
employing the developed (12), (14), (18) and (26).

3.3 Current loop references

Since the fundamental positive-sequence components of the
non-linear load currents appear as D C quantities in the
synchronous reference frame rotating at the frequency o,
they can be easily separated from the load actual currents.
As shown in Fig. 2a, the load currents iL1and iL2are sensed " 60 300 420 660 780 10201140 13801500

and transformed into iw and iLy in the synchronous

reference frame. The d-axis signal iLd is then processed
through a fourth order Butterworth low-pass filter in order
to isolate its DC quantity and then extract its harmonic
component ik,h. The low-pass filter has a pass band
fp = 20 Hz, and four poles having the same frequency
fo=25Hz. The q-axis current i k is inverted and then
used as a harmonic and reactive reference component i8. 60 300 420 660 780 I0201140 13801500
frequency [Hz]
Further, t h s method requires the generation of in-phase
and in-quadrature sinusoidal synchronous signals with the Fig. 3 Harmonic spectrum of iL (top) and i, (bottom)
phase voltage U,, which is obtained by means of a PLL
circuit.
The current loop references are then obtained as 4.2 Performance and dynamic response of the
follows: current controller
The d-axis current id, the q-axis current iq and the phase 'a'
current il of the active filter in the steady state operation are
superposed with their respective references and shown in
Fig. 4. The results highlight that the filter-injected oscillating
current harmonics track their reference templates with high
levels of accuracy.
4 Simulation results

In order to validate the accuracy of the proposed controller,

the system was simulated using the 'power system
blockset' in Matlab/Simulink environment. The system ca
parameters used in these simulations are shown in 0 -21
Table 1. The simulation goals are to examine the 0.068 0.072 0.076 0.08 0.084
capability of the controller to fulfil the following four
different criteria:
Low order current harmonic compensation (below 1
kHz).
0.068 0.072 0.076 0.08 0.084
Performance and dynamic response of the current
controller.
Multi-functional performance.
Compensation of load imbalances.
I

The results of each goal are shown separately as follows. 0.068 0.072 0.076 0.08 0.084
time [SI

Fig. 4 Tracking performance of the inner loop (from top to

Table 1: System parameters bottom): id iq, and il with their references

Line voltage, and frequency 11OV (rms), 60Hz Further, the nonlinear load was subjected to 100% step
Line inductance L,= 0.05 mH increase and step decrease by the same ratio at t = 92 ms
Active filter parameters L,= 1 mH, and t = 150ms, respectively. The firing delay angle of the
R,= 0.1 R, thyristors is fixed at 20". Both the current harmonics and
C= 1000 pF the power factor compensations are active during this test.
Active filter DC bus voltage Vdc=350V The relevant waveforms shown in Fig. 5a and 5b illustrate
Nonlinear load output power PL=2kW
the dynamic response of the active filter to these changes. It
is important to note the effect of the controller on id when
Controller frequencies radls.
wni= 6000~
U& transiently changes from its set point value, whereas the
CO,,= 8On radls
supply voltage and current remain in phase.
IEE Proc-Electr. Power Appl., Vol. 150, No. 4, July 2003 377
 -- 0.06 0.10 0.14 0.18 0.22

0.06 0.10 0.14 0.18

b
._ x I
0.06 0.10 0.14 0.18 0.22
.-
-5
,
0.06 0.10 0.14 0.18

-5I
-~ ~
0.06 0.10 0.14 0.18
0.06 0.10 0.14 0.18 0.22
time [SI
a P

-I
$ 0
41

."41 I
0.06 0.10
time [SI
Fig. 6 Multifunctional capabilities as demonstrated by the wave-
forms ($-om top to bottom): (iAl,VI), il, i'h iq and v,Ic
0.14 0.18

0.06 0.10 0.14 0.18 0.22

I
0.06 0.10 0.14 0.18 0.22

360 1
0.20 0.25 0.30 0.35

I
0.04 0.06 0.08 0.10 0.12
-5
time [SI
b 0.20 0.25 0.30 0.35

Fig. 5 5 1
a From top to bottom: {iLl, v,}, {is],U , } and il resulting from step
changes in the nonlinear load
b From top to bottom: iq,and vd,. resulting from step changes in the
nonlinear load I I I I
0.20 0.25 0.30 0.35
time [SI
Fig. 7 Imbalanced load currents
4.3 Multi-functional performance
The capability of the active filter towards the current
harrnonics elimination with or without reactive power
compensation is tested (Fig. 6). Before the instant t = 70 ms,
the compensation features were turned off while the DC
voltage level was maintained at its set point. The small
fundamental current flowing into the active filter during t h s
interval is in phase with phase voltage and compensates for I I I I

0.20 0.25 0.30 0.35

losses in the active filter (losses due to R, and the switch
resistors). At time t = 70 ms, only the harmonics compensa-
tion feature is turned on, and consequently the supply
current becomes sinusoidal but still phase shfted with the
voltage. At t = 120ms, the reactive power compensation is
activated, and the supply current then becomes locked with I I I
0.20 0.25 0.30 0.35
the phase voltage.

4.4 Compensation of load imbalances

This test aims to evaluate the active filter capability to
compensate non-characteristic current harmonics. To carry
out the test, a single-phase thyristor controlled rectifier I I I

0.20 0.25 0.30 0.35

operated with 40" phase shifting is connected in parallel time [SI
with the 6-thyristor bridge nonlinear load. The single-phase
rectifier is connected between phases '1' and '2' by closing Fig. 8 Supply currents with imbalanced load

378 IEE Proc.-Electr. Power Appl.. Vol. 150, No. 4, July 2003
the switch SW as shown in Fig. 2a. The resultant closed loop error dynamics of current trachng and voltage
unbalanced three-phase load currents are shown in Fig. 7. regulation. The simulation results prove that the following
The shaped supply current waveforms are illustrated in objectives have been successfully achieved: (i) drastic
Fig. 8. It is noted that these latter currents are sinusoidal reduction of the current harmonics in the supply mains;
and balanced. The spectral analysis of load and line currents (ii) good performance of the closed loop response of the
depicted in Fig. 9 shows the ability of the active filter to active filter under both dynamic and steady state opera-
cancel the non-characteristic harmonics. tions; (iii) multi-functional capabilities of the active filter,
and; (iv) compensation of load imbalances.

1.o
F
6 Acknowledgments
5 IL11(Arms): 4.9644 Is1 1(Arms): 3.985
2 0.5 The authors wish to’ thank the Canada Research Chair in
-
5
Electric Energy Conversion and Power Electronics and the
0 National Research and Engineering Council of Canada for
60 300 660 1020 1500 60 300 660 1020 1500
their financial support.

IL21(Arms): 4.8662 Is21(Arms): 3.8331

References
Magi, H.: ‘Trends in active power line conditioners’, IEEE Trans.
Power Electron., 1994, 9, (3), pp. 263-268
60 300 660 1020 1500 60 300 660 1020 1500 Singh, B., AI-Haddad, K., and Chandra, A.: ‘A review of active filters
for power quality improvement’, IEEE Trans. Ind Electron., 1999, 46,
(S),pp. 96&971-
IL31(Arms): 2.3464 Is31(Arms): 3.9395 Saetieo, S., Devaraj, R., and Torrey, D.A.: ‘The Design and
Z 0.5 5 0.5 Implementation of a Three-phase Active Power Filter Based on Sliding
a Mode Control’, IEEE Trans. Ind. Appl,, 1995, 31, (5), pp. 993-1000 -
Guffon, S.. Toledo, AS., Bacha, S., and Bomard, G.: ‘Indirect sliding
mode control of a three-nhase active Dower filter’. Proc. IEEE Power
60 300 660 1020 1500 60 300 660 1020 1500 electronics specialists Conf. (PESC), Fukuoka, Japan,
17-22 May 1998, pp. 1408-1414
Fig. 9 Load und supply curreni yectmms with irnbulanced load Bhattacharya, S., Veltman, A., Divan, D.M., and Lorenz, R.D.: ‘Flux
based active filter controller’. Proc. Meeting of IEEE Industry
Applications Society (IAS), Orlando, FL, 8-12 October 1995, pp.
-7dRL7dQ
- . , I~
. Y _

5 Conclusions Marques, G.D., and Verdelho, P.: ‘Control of an active power filter
based on input-output linearization’. Proc. IEEE Industrial electronics
Conf. (IECON), Aachen, Gemany, 31 August4 September 1998, pp.
In this paper, the exact feedback linearkation control has 456-46 1
been successfully applied to design a shunt active power Mendalek, N., Fnaiech, F., AI-Haddad, K., and Dessaint, L.-A.: ‘A
filter controller. The application of the control strategy is Non-Linear Optimal Predictive Control of a Shunt Active Power
Filter’. Proc. IEEE Industry Applications Society Meeting (IAS),
possible because the nonlinear state-space active filter model Pittsburg, PA, USA, 13-18 October 2002, pp. 7C-77
is divided into two separate loops. Therefore, the exact Mendalek, N., AI-Haddad, K., Dessaint, L.A., and Fnaiech, F.:
‘Nonlinear Control Strategy Applied to a Shunt Active Power Filter’.
feedback linearkation control is proposed for both loops. Proc. IEEE Power electronics specialist conf. (PESC), Vancouver,
Then, the pole placement strategy is used to synthesise the Canada, June 2001, pp. 1877-1882

IEE Proc.-Electr. Power Appl, Vol. 150, No. 4, July 2003 319