IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO.

6, JUNE 2015 3351

Optimization Algorithm for Selective

Compensation in a Shunt Active Power Filter
Jos Carlos Alfonso-Gil, Emilio Prez, C. Ario, and Hector Beltran

AbstractThis paper proposes an optimization algo-

rithm based on linear matrix inequalities (LMIs) to be im-
plemented in a shunt active power lter (SAPF) enabling it
to perform a selective compensation of nonefcient powers
when the rated power of the SAPF is limited. The sys-
tem uses the IEEE Standard 1459 to identify the different
power terms (active, reactive, unbalance, and harmonics)
demanded by local loads in a three-phase four-wire sys-
tem. Then, the algorithm solves a quadratically constrained
quadratic program by means of an LMI formulation and
calculates the optimal reference currents that allow the Fig. 1. Block diagram of a SAPF connected to the grid.
SAPF to cancel a selection of the previously identied
nonefcient power terms. For this selection, the algorithm
presents different weighting coefcients in the cost in- strategies [1], [2] based each on a different electric power the-
dex to be optimized. These coefcients assign a rela- ory. Similarly, multiple current control techniques [3][8] and
tive importance to each different nonefcient power term, grid synchronization methods [9], [10] have been developed in
giving priority to some over the others. Moreover, the the last years. Focusing on the compensation strategies, active
optimization performed in this work to calculate the refer- filters are typically addressed as global filters [2], [6], [11],
ence currents considers that the SAPF presents a power
limitation. Finally, the reference currents tracking has been [12] where the nominal power of the system is large enough
implemented using a proportional feedforward controller so as to compensate all the nonefficient phenomena (reactive,
which assures a constant commutation frequency of the unbalance, and harmonic distortion) caused by the surrounding
SAPF. Experimental results demonstrate the validity of the loads. However, when the nonefficient powers of the load
proposal. vary significantly (due to load expansions or the simultaneous
Index TermsActive power lter, IEEE Std. 1459, linear operation of loads), it may be difficult to preview the size
matrix inequality (LMI) algorithm, selective compensation. of the SAPF to compensate simultaneously all the potential
nonefficient powers. In this context, it is more convenient to
I. I NTRODUCTION
compensate separately the different nonefficient phenomena or

S HUNT active power filters (SAPFs) have traditionally been

used to improve the quality of the electric systems, reduc-
ing the effects of reactive, unbalanced, and harmonic distortion
phenomena produced by nonefficient loads. To that effect, the
SAPF analyzes the grid and generates the nonefficient currents
demanded by those systems connected downstream of its point
of common coupling (PCC), as shown in Fig. 1. In this way, the
transport and distribution networks only have to deliver useful
power, thus improving the quality and efficiency of the electric
system.
The continuous development of the SAPF technology has
favored the emergence of various nonefficient compensation
Manuscript received February 15, 2014; revised May 5, 2014 and
July 17, 2014; accepted November 7, 2014. Date of publication De-
cember 9, 2014; date of current version May 8, 2015. This work
was included within the project Development and implementation of
control strategies for a compensator-generator system in Smart Grid
applications, with reference P11A2011-12 supported by the Universitat
Jaume I de Castell and the Fundaci Caixa Castell-Bancaixa.
The authors are with the Industrial Engineering and Design Depart-
ment, Universitat Jaume I, 12071 Castellon de la Plana, Spain (e-mail:
jalfonso@uji.es; pereze@uji.es; arino@uji.es; hbeltran@uji.es).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TIE.2014.2378751
0278-0046 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
even a combination of them. Thus, some authors propose in
[4] a selective compensation of harmonic load currents using
a set of PI controllers tuned at each of the frequencies of the
harmonics to be eliminated. A different approach is presented
in [8] where the authors also propose a selective compensation
of harmonics with a control technique which they call selective
harmonic compensation. Both approaches in [4] and [8] do
not face either the reactive or the unbalance compensation, and
as the number of harmonics to be compensated increases, the
proposal in [4] may prove unviable. One step beyond, other
authors propose in [13] a selective compensation strategy which
does include all nonefficient phenomena. This technique is
based on the compensation theory defined in [1], and it uses
a neural network to calculate the reference currents. The same
authors propose in [14] an algorithm that assigns priority to
the compensation of harmonics in front of other nonefficient
phenomena when the filter presents current limitations.
Nonetheless, it does not allow to perfectly profit all the nomi-
nal power of the filter. Furthermore, the authors use in both [13]
and [14] a hysteresis band regulator to manage the compensa-
tion currents. This means that the switching frequency of the
converter is dependent on the load parameters and on the grid
voltage [15], causing, in this way, an uncontrolled harmonic
3352 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 6, JUNE 2015

distortion in the filter output current. On another level, other
authors present in [16] a selective compensation using the IEEE
Std. 1459 [17] to identify each of the nonefficient power terms
to be compensated. The authors use in that work a proportional
current regulator and an SVPWM modulation, keeping con-
stant, in that way, the switching frequency of the filter. Also, in
[18], the same authors propose another selective compensation
strategy, but in this case, they generate the compensation cur-
rent references using the method of the equivalent conductance.
Both strategies proposed in [16] and [18] perform an individual
or a combined compensation of the nonefficient powers, but
none of them take into account the limitations of power that
the filter can experience during compensation.
This paper proposes an optimization algorithm based on
linear matrix inequalities (LMIs) which, using the IEEE Std.
Dealing with Se1 , it can be divided according to (5) by using
the Fortescue transformation. Therefore, it is formed by its
positive-sequence component (S1+ ) and a second component
including not only the negative and the zero sequences but also
the crossed products of the fundamental currents and voltages
of any sequence, (SU 1 ). The latter is usually associated to the
unbalance of the system and is called unbalance power
 2
2
Se1 = S1+ + SU2 1 . (5)
In the same way, the term S1+ contains another two terms: a
phase and a quadrature one, as shown in (6). These correspond
to the fundamental positive-sequence active power (P1+ ), which
represents the transference of the actual useful energy, and to
the fundamental positive-sequence reactive power (Q+ 1 ), which

1459 to identify the different power terms, enables a selective represents the bidirectional energy flow due to the phase shift
compensation of the various nonefficient powers existing in between voltages and currents
   2
the grid without exceeding the nominal power of the SAPF. 2
Moreover, this paper presents a proportional feedforward cur- S1+ = P1+ + Q+ 1 . (6)
rent regulator designed to control the selected and defined Moreover, some authors propose the decomposition of SU 1
compensation currents while keeping constant the switching into three different terms [19]. In order to derive these new
frequency of SAPF. terms, Ve1 and Ie1 , both introduced in (2), are divided according
This paper is organized as follows. Section II introduces to (7) and (8), respectively,
the theoretical foundations of the algorithm used to analyze
 2
nonefficient phenomena. In Section III, the reference current 2
Ve1 = V1+ + (VU 1 )2 (7)
generation using the LMI strategy is described. Then, the SAPF 2
 + 2 2
Ie1 = I1 + IU 1 (8)
characteristics and the controller design are presented in Sec-
tion IV. Section V is devoted to presenting some compensation where V1+ and I1+ are the fundamental positive-sequence volt-
results. Finally, some conclusions are stated in Section VI. age and current and VU 1 and IU 1 are the fundamental unbalance
voltage and current, respectively. From (5) and taking into
II. T HEORETICAL B ASIS FOR THE G ENERATION account that S1+ can be expressed by the product of I1+ and
OF THE C OMPENSATION C URRENTS
V1+ , (9) is deduced
 2
A. Breakdown of Power Terms According to IEEE SU2 1 = Se1
2
3V1+ I1+ . (9)
Std. 1459
Then, if I1+ and V1+ are isolated in (7) and (8) and substituted
The correct identification of the different power terms form- in (9), the latter can be rewritten as
ing the apparent power flowing through a line is the basis to  2  2 
perform an optimal compensation of the nonefficient powers. SU2 1 = Se1
2
9 Ve1 VU21 Ie1 IU2 1 . (10)
The IEEE Std. 1459 defines the effective apparent power (Se )
as in (1), where Ve1 and Ie1 are the fundamental components Setting Se1 as a function of Ve1 and Ie1 , (11) is obtained
of the effective voltage and current and VeH and IeH are
SU2 1 = (3Ve1 IU 1 )2 + (3VU 1 Ie1 )2 (3VU 1 IU 1 )2 . (11)
the nonfundamental components of the effective voltage and
current, respectively. Thus, according to it, Se includes all the The aforementioned terms can be expressed by means of
power terms concerning efficient and nonefficient phenomena their symmetrical components. Hence, by using the Fortescue
that could be required by a generic three-phase load transformation, Ie1 is formed by the currents in
   2
Se2 = (3Ve1 Ie1 )2 +(3Ve1 IeH )2 +(3VeH Ie1 )2 +(3VeH IeH )2 . 2
I1+ + I1 + 4 (I10 )
2
Ie1 = (12)
(1)
The first term belongs to the fundamental effective apparent where I1 and I10 are the fundamental negative- and zero-
power (Se1 ), defined as in (2), while the other terms belong to sequence components of current. On the other hand, using (8)
the nonfundamental effective apparent power (SeN ), defined as and (12), the currents included in IU 1 are defined in
in (3). Hence, (1) can be expressed as in (4)  2  2
IU2 1 = I1 + 4 I10 . (13)
2
Se1 = (3Ve1 Ie1 )2 (2)
 Therefore, with (11) and (13), one can write the terms in SU 1
SeN = 9Ve1 2 I 2 + 9V 2 I 2 + 9V 2 I 2 (3)
eH eH e1 eH eH as in
 2  2  0 2
Se2 = Se1
2 2
+ SeN . (4) SU2 1 = 9VU21 I1+ + 9Ve1 2
I1 + 36Ve1 2
I1 . (14)
ALFONSO-GIL et al.: OPTIMIZATION ALGORITHM FOR SELECTIVE COMPENSATION IN SAPF 3353

index, using an auxiliary variable as in

 1 
P0 x
r0 xT P0 x > 0 > 0. (21)
xT r0

Thereupon, the LMI problem to be solved is

min r0 subject to :
Fig. 2. Resolution of the Se used to obtain the compensation currents x
of SAPF.  1 
Pi x
> 0, i = 0 . . . p. (22)
xT ri
Regarding the nonfundamental effective apparent power
(SeN ), it is also divided into three terms: the current distortion Thus, this optimization problem is resolvable using well-
effective power (DeI ), the voltage distortion effective power known LMI solving tools such as Yalmip [22] or Sedumi [23].
(DeV ), and the harmonic effective apparent power (SeH ).
These are defined in (15)(17), respectively,
III. R EFERENCE C URRENT G ENERATION
DeI = 3Ve1 IeH (15) The generation of reference currents to introduce in the
SAPF is determined by the magnitude of the various nonef-
DeV = 3VeH Ie1 (16)
ficient powers that would be desirable to cancel but is also
SeH = 3VeH IeH . (17) limited by the nominal power of the SAPF itself. Measuring
the load currents and using the power definitions provided by
Thereupon, SeN quantifies the harmonic effective power the IEEE Std. 1459, the reference currents are calculated so as
consumed by the loads, which can be also quantified by to selectively compensate the different nonefficient phenomena
without exceeding the maximum rated current of the SAPF. To
IeH do this, a cost index to be minimized is formulated. This index
T HDeI = . (18)
Ie1 contains all the power terms to be compensated and assigns a
relative weight to each of them. It also takes into account the
Finally, Fig. 2 summarizes the different terms obtained from power limitation of SAPF. For the proper calculation of this
the Se decomposition. index, it is necessary to correctly identify the grid, the load, and
The selective compensation developed in this paper intends the SAPF phase currents.
to cancel the different nonefficient power terms (Q+1 , Su1 , and In this sense, (23) defines each of the SAPF phase currents
SeN ) separately or even a combination of them. (k = a, b, c) in rectangular coordinates, where the subscripts r
and i denote the real and the imaginary part, respectively
B. Quadratically Constrained QP Ik1 SAPF = Ik1r SAPF + jIk1i SAPF . (23)
Once the various nonefficient power terms are identified,
some tool is required in order to generate the optimal com- On the other hand, (24) defines the rms value of the SAPF
pensation current references to introduce in the SAPF. In this phase currents
case, this work uses the quadratically constrained quadratic
programming (QP) and, more precisely, the transformation of Ik2 SAPF = Ik1
2 2
SAPF + IkH SAPF (24)
an optimization problem into a LMI problem, which is an
where Ik SAPF is the rms value of the total current for each of
efficient and elegant way of solving this type of challenges.
the SAPF phases and IkH SAPF is the rms value of the total
To do that, note the simplification of a generic quadratically
harmonic content existing in those phase currents.
constrained QP problem introduced in [20] and defined in
In the same way, (25)(28) define the fundamental and the
rms currents for both load and grid phases, respectively
min xT P0 x subject to
x
Ik1 load = Ik1r load + jIk1i load (25)
xT Pi x ri , i = 1 . . . p. (19)
Ik2 load = Ik1
2 2
load + IkH load (26)
If the different Pi matrices are positive definite, the Schur
complement [21], which allows transforming the quadratic con- Ik1 s = Ik1r s + jIk1i s (27)
straints into LMIs, can be applied. This transformation yields
 1  Ik2 s = Ik1
2 2
s + IkH s . (28)
Pi x
ri x Pi x > 0
T
>0 (20)
xT ri According to the sign criteria defined in Fig. 1, the resulting
current at the PCC can be written as in
where the inequality sign indicates that the matrix is positive
definite. The same transformation can be applied to the cost Ik1 load = Ik1 SAPF + Ik1 s . (29)
3354 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 6, JUNE 2015

Finally, it is to be noted that, provided that all the harmonic D. Apparent Power Due to Current Harmonics
currents have been included within a single term, it makes no
The current harmonic evaluation is global, i.e., without dis-
sense to operate with current phasors and it is preferable to use
tinguishing the harmonic order, as defined in the IEEE Std.
the rms values instead.
1459. Therefore, taking again into account the current signs
defined in Fig. 1 at the PCC, (33) can be written
A. Constraints
ikH load (t) = ikH SAPF (t) + ikH s (t). (33)
Once the different phase currents have been defined, the
constraints of the problem have to be set. In this paper, the main To the same extent, (34) defines the value of the harmonics
constraint introduced is associated to the limitation in the rms introduced by the SAPF as a ratio of the load harmonics
total phase-current value defined for the SAPF, i.e.,
ikH SAPF (t) = ikH load (t) (34)
Ik SAPF Imax where is a variable defined between 0 and 1 which will
determine the harmonic compensation level for each of the
where Imax is usually defined as the rms SAPF nominal current. phases. In this way, it can be guaranteed that the effective
This inequality can also be rewritten as in harmonic current value will be as in
2
Ik1r 2 2
SAPF + Ik1i SAPF + IkH SAPF Imax
2
. (30) IeH SAPF = IeH load . (35)

Therefore, due to the scalar nature of the coefficient, the

definition in (36) can be granted
B. Unbalances in the Fundamental Component
IeH s = (1 )IeH load . (36)
The obtainment of the positive-, negative-, and zero-
sequence grid current components requires the use of the Moreover, according to (36), the SeN terms defined in
Fortescue transformation. As described in Section II-A, the use (15)(17) can be reformulated as in
of these components allows defining the term SU 1 , defined in
(14), that minimizes the power term associated to unbalances. DeI = 3Ve1 IeH load (1 ) (37)
In case compensating it was the unique goal of the SAPF, the DeV = 3VeH Ie1 s (38)
reference currents could be calculated, using the measured load
currents and grid voltages, according to SeH = 3VeH IeH load (1 ). (39)

Note from these equations that the DeV straightly depends

min SU2 1 subject to on the fundamental grid current component and on the grid
SAPF + Iki SAPF + IkH SAPF Imax
2
Ikr 2 2 2 voltage harmonics. Given that the SAPF controls its currents
+ and, consequently, the grid currents, the only way to limit
I1s 1 a a2 Ia1s this nonefficient power term is to reduce the fundamental grid
I1s
1
= 1 a2 a Ib1s current component. In this way, because the grid voltage is
3
0
I1s 1 1 1 Ic1s independent of the system, the optimization performed in this
Ik1 load = Ik1 SAPF + Ik1 s (31) work does not take into account this term.

where a = e(2/3)j . One can observe how this problem con-
tains complex variables and quadratic constraints that require
some treatment to become a resolvable LMI problem using the
tools introduced in Section II-B. This transformation is detailed
in the coming sections.
C. Fundamental and Positive-Sequence Reactive Power
To improve the power factor of the system, the fundamental
and positive-sequence reactive power can be calculated as in
E. Active Power Delivered by the SAPF
Finally, the SAPF could also control the active power ex-
changed with the grid. This power is set to a null value if the dc
bus of the SAPF is formed by capacitors, given that any active
power exchange would modify the bus voltage dramatically.
Conversely, if the dc bus is formed by an energy storage sys-
tem presenting a significant energy capacity (batteries, super-
caps, etc.), some control strategies can be implemented in the
SAPF to also manage the fundamental active power exchange
(Pref SAPF ). This power can be written as a function of the
    optimization variables as in
Q+ + I+ .
1 = 3 Im V1 1 (32)  

Pk1 SAPF = Re Vk1 s Ik1 SAPF (40)
Therefore, if the unique goal of the SAPF was, in this case, Pa1 SAPF + Pb1 SAPF + Pc1 SAPF = Pref (41)
SAPF
to minimize this power, Q+ 1 , a new optimization problem that
used the same constraints as that in (31), could be defined, but where Vk1 s is the fundamental voltage for each of the grid
this would incorporate (Q+ 2
1 ) as the cost index instead. phases.
ALFONSO-GIL et al.: OPTIMIZATION ALGORITHM FOR SELECTIVE COMPENSATION IN SAPF 3355

F. Optimization Problem As described in that section, the inequality can be converted

to an LMI form
On the basis of the previous discussion, if all the noneffi-

cient power terms are to be considered, the resolution of the 1 0 0 Ikr SAPF
reference currents can be defined as an optimization problem 0 1 0 Iki SAPF
> 0.
in the variables IkrSAPF , IkiSAPF , and when Ikrload , Ikiload , 0 0 I2
1

kH load
IeHload , V1rs , V1is , and VeHs are available. Thus Ikr SAPF Iki SAPF 2
Imax
 2  2 
min KQ Q+ 1 + KU SU2 1 + KH DeI 2
+ SeH Finally, the optimization index is also quadratic, and there-
subject to fore, it has to be rewritten as an LMI. First, the SU 1 term is
replaced by (47)
SAPF + Iki SAPF + IkH load Imax   0 2 
2 2 2 2 2
Ikr (42)  2  2  2
Pa1 SAPF + Pb1 SAPF + Pc1 SAPF = Pref SAPF (43) KQ Q + 1 + KU 9VU21 I1+ + 9Ve1 2
I1 + 36Ve1 2
I1
 2 2

where +KH DeI + SeH .
Ik1 load

s
= Ik1SAPF + Ik12 (44)
Next, it is reformulated as an inequality of the form (21).
I1 s
+
1 a a Ia1 s
I1s = 1 1 a2 a Ib1 s To do so, each of the current terms is separated into the real and
(45)
3 imaginary parts as in (I+ 2 + 2 + 2
1 ) = (I1r ) + (I1i ) , and vector x and
I10 s 1 1 1 Ic1 s
    matrix P0 are defined
Q+ + +
1 = 3 Im V1s I1s (46)
      x = [Q+ + + 0 0
1 I1 r I1 i I1 r I1 i I1 r I1 i De1 SeH ]
T
2 2
I1 + 36Ve1
2
SU2 1 = 9VU21 I1+ + 9Ve1 2 2
I10 (47) 
DeI = 3Ve1 IeH load (1 ) (48) P0 = diag KQ , 9KU VU21 , 9KU VU21 , 9KU Ve12 2
, 9KU Ve1 ,

SeH = 3VeH  IeH load (1 )
 (49) 2
36KU Ve1 2
, 36KU Ve1 , KH , KH

Pk1SAPC = Re Vk1 s Ik1 SAPC . (50)
leading to the LMI
Note that the inclusion of the different weighting coefficients  
in the cost index (KQ weights the phase shift, KU weights the P01 x
> 0.
unbalance, and KH weights the harmonic distortion) will allow xT r0
to assign a relative importance to each different nonefficient
This LMI forces the optimization index to be lower than an
power term.
auxiliary variable r0 , which will be the variable to be minimized
The previous optimization problem is still not in the proper
as in (21).
form to be solved by an LMI optimization tool. Therefore, the
The LMI optimization provides the optimal values of
equations will be reformulated to the form (20) and (21).
Ikr SAPF , Iki SAPF , and . In order to use these results as ref-
First, we consider (43)(46) and (48)(50). In all of them,
erences to the current controller described in the following sec-
the optimization variables appear as a linear function of their
tion, they have to be expressed in the time domain as follows:
real and imaginary parts. This allows the direct inclusion in the

LMI formulation. For brevity, only the first of them is detailed 2 2
Ik1 SAPF = Ikr SAPF + Iki SAPF (51)
as follows:
Sk1 SAPF = (Vkr s + jVki s )(Ikr s jIki s ) k1 SAPF = tan1
Iki SAPF
(52)
Pk1 SAPF = Vkr s Ikr s + Vki s Iki s . Ikr SAPF

This way, the total active power constraint results ik ref (t) = 2Ik1 SAPF sin(100t + k1 SAPF )


Vkr s Ikr s + Vki s Iki s = Pref SAPF . + ikH load (t) (53)
k(a,b,c)
where ikH load (t) is the previously measured load harmonic
Note that the obtained equation can be expressed in the current used for the optimization.
optimization variables by using (44).
Next, constraint (42) is considered. As the optimization
variables appear in quadratic form, this inequality has to be IV. D ESCRIPTION OF THE S YSTEM
dealt with as described in Section II-B. First, the equation is A. SAPF
written as in (19)
The proposal presented in this paper has been tested in a
1 0 0 system such as the one represented in Fig. 3, where the SAPF
[ Ikr SAPF Iki SAPF ] 0 1 0 is connected at the PCC with the low voltage grid and some
2
0 0 IkH load local loads. As can be observed, the SAPF is formed by a three-
Ikr SAPF legged split-capacitor inverter and a three-phase output filter
Iki SAPF Imax2
. (with values Ra = Rb = Rc = 0.4 and La = Lb = Lc =
6 mH). Regarding the load, used in both the simulations and the
3356
Fig. 3. Schema of connection of the complete experimental system.
laboratory tests, this contains a linear unbalanced three-phase
load in parallel to a nonlinear balanced load. The former is
constituted by resistances and windings (with values Raload =
32 , Laload = 18 mH, Rbload = 50 , Lbload = 9 mH, and
Rcload = 100 ), and the latter is constituted by three one-
phase noncontrolled rectifiers feeding an RC load through an
inductive filter (with values L = 5 mH, C = 2200 F, and
R = 100 ).
The capacitance of the dc bus is C1 = C2 = 5.44 mF, and
the operating voltage is 550 V. Its midpoint is connected to
the grid neutral to allow the circulation of the zero-sequence
currents. To protect the dc bus against overvoltage, the SBK
IGBT transistor and external RBK are used. The control of the
dc bus voltage implements a regulator such as that proposed
in [16] and [18]. A 400/200-V wye-wye transformer connects
the SAPF and the three-phase loads to the grid in order to get
a good modulation gain from the 550 V of the dc bus. The
switching frequency is 27.15 kHz, which is also the sampling
frequency implemented for the control and the acquisition of
currents and voltages. In this way, the load current is sampled
543 times every grid cycle.
Thereupon, once a second, the DSP driving the SAPF sends
the last-cycle load currents via a RS232 communication to a PC.
This calculates the updated current references and returns them
to the DSP, which generates the SAPF modulation according to
them. Thus, an external PC is the one running the optimization
algorithm. With this approach, it is guaranteed that the SAPF
will not exceed its nominal power. On the other hand, when a
load current transient occurs, the compensation currents are not
adequate, for 1 s, as the new references are still being calculated
on the PC.
B. Current Controller
A current regulator quick enough to correctly track the cur-
rent references and thus properly perform the nonefficient phe-
nomenon compensation has been designed and implemented.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 6, JUNE 2015
Fig. 4. Block diagram of the current control implemented in the SAPF.
Fig. 5. Block diagram of the feedforward current regulator.
Fig. 4 depicts the block diagram of this current controller.
This proposed regulator is a proportional (P) feedforward
controller such as the one described in [24] (see Fig. 5). It
is formed by two terms: a standard P regulator (which gen-
erates a control action proportional to the tracking error) and
a feedforward term, which inverts the system dynamics. To
do this inversion, the grid voltage and current reference are
needed. This way, in an ideal situation with no disturbances,
the second term calculates the voltages needed to get the desired
reference current from the SAPF. This control structure allows
an improved dynamic response from a standard P regulator in
systems, such as the one described here, where the feedforward
term only needs to calculate derivatives of a reference signal
(and not a measurement). Usual problems in feedforward con-
trollers arising from the fact of differentiating noisy measured
signals are, in this way, avoided. Also, note that a saturation
block has been added in order to limit the maximum pointwise
value of the reference current. This way, references higher than
the peak current of the IGBT transistors are never demanded.
To design the feedforward regulator (see Fig. 5), the SAPF
output current and voltages are formulated as in
d
vk SAPF = Rk ik ref +Lk ik ref +vk s +P (ikref ik SAPF )
dt
(54)
where vk SAPF is the converter output voltage, vks is the grid
voltage, and Rk and Lk are the values of the SAPF output filter
impedance of each phase. Then, the P regulator is defined by
forcing the settling time to tst = 2 ms and kP W M = 1, using
the characteristic (55). In this way, a proportional gain such as
that in (56) is obtained [25]
1
1 + P GoL (s) = 1 + P kP W M = 0 (55)
R k + Lk s

|Rk + Lk s| 
|P | = = 11.6. (56)
kP W M  s=4/tst
ALFONSO-GIL et al.: OPTIMIZATION ALGORITHM FOR SELECTIVE COMPENSATION IN SAPF
Fig. 6. Grid and SAPF neutral current transient during the connection
of the SAPF for a global compensation (simulated results).
V. O BTAINED R ESULTS
Both simulations and experimental tests have been carried
out to analyze the performance of the proposed optimiza-
tion algorithm. Simulations used a digital model developed
in Matlab/Simulink and its SimPowerSys toolbox. Likewise,
the experimental tests used a small-scale SAPF prototype de-
veloped in the laboratory. The control of this prototype was
implemented in a TMS320F28335 DSP from Texas In-
struments, and the power stage used an intelligent power
module containing IGBT transistors (PM150RL1A120) from
Mitsubishi Electric. The rest of the components and operating
characteristics are those defined in Section IV-A. For such a
scenario, the performance of the SAPF under different op-
eration conditions has been analyzed. The analysis focuses
on the evaluation with Mathcad of the different power terms
associated to each of the corresponding nonefficient phenomena
described by the IEEE Std. 1459. The evaluation is performed
both before and after activating the compensation, using the
voltages and currents measured at the PCC. The study cases
ranged from a global compensation of the nonefficient phe-
nomena to different selective compensations that prioritized the
various nonefficient power terms. Finally, the performance of
the proposed optimization algorithm is compared to a similar
approach from the literature. For the first case, the SAPF is
supposed to present enough power capacity to compensate all
the nonefficient phenomena generated by local loads. Figs. 6
and 7 show the simulated and experimental results, respec-
tively, obtained during the connection transient of this global
compensation. The scale used in the oscilloscope probes is
10 mV/A in all experimental cases. Both figures demonstrate
how unbalances, phase shift, and harmonic distortion are almost
entirely canceled from the grid currents after the connection
of the SAPF. Likewise, note how the zero-sequence current
flowing through the grid neutral before the compensation is
driven afterward by the SAPF neutral.
Table I shows the values of the various nonefficient power
terms experimentally measured before and after activating this
global compensation. These values confirm the comments an-
ticipated from Figs. 6 and 7. In this regard, it is remarkable
3357
Fig. 7. Grid and SAPF neutral current transient during the connection
of the SAPF for a global compensation (experimental results).
TABLE I
P OWER T ERMS W HEN THE SAPF H AS N O P OWER C ONSTRAINTS
how Su1 , Q+ 1 , and SeN have been decreased by 88.42%,
95.33%, and 79.22%, respectively. Also, note that, in this case,
the values assigned to the weighting coefficients are irrelevant
because the SAPF has enough power.
All the other cases analyzed comprehend selective com-
pensation scenarios. These imply a power limitation of the
SAPF capacity introduced as a limit on its phase current. The
maximum current tolerated in the following experimental tests
is set to 1.4 A rms, in accordance with the loading conditions
available in our laboratory. However, the different selective
cases vary in the values defined for the weighting coefficients
(KU = KQ = KH ), assigning priority with those 1.4 A to the
compensation of different nonefficient phenomena every time.
The experimental results for some of these cases are presented
in the following. The first one assigns the same importance
to the three coefficients KU = KQ = KH = 1. Fig. 8 shows
the resulting currents demanded to the grid once the SAPF
has started compensating the three nonefficient phenomena.
Note the difference with the global compensation previously
introduced in Fig. 7 due to the current constraints defined
for the SAPF. Table II summarizes the exact values of the
different power terms resulting from this compensation. Also,
for this case, Fig. 9 shows the reference currents generated by
the algorithm and the output currents of the SAPF during the
compensation. Observe that the proposed controller achieves a
satisfactory tracking of the reference currents for all the three
phases.
Also note the rms phase-current values Ia SAPF = 1.39 A,
Ib SAPF = 0.99 A, and Ic SAPF = 1.40 A. As expected, these
are below 1.4 A, being well below the Ib SAPF value. This is
because the harmonics are compensated equally in all three
3358
Fig. 8. Grid currents after compensation with KU = KQ = KH = 1.
TABLE II
P OWER T ERMS OF THE G RID C URRENTS A FTER S ELECTIVE
C OMPENSATION (SAPF O UTPUT C URRENT L IMITED TO 1.4 A RMS )
Fig. 9. Output versus reference currents of the SAPF with KU =
KQ = KH = 1.
phases (36). Then, once the current has reached the maximum
in one phase, this one fixes the harmonic compensation limit in
the other two, avoiding causing an unbalance in the harmonics.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 6, JUNE 2015
Fig. 10. Grid currents after compensation with KU = 1000, KQ =
1000, and KH = 1.
Fig. 11. Grid currents after compensation with KU = 1, KQ = 1, and
KH = 1000.
The second selective case analyzed assigns greater priority to
the compensation of unbalances and phase shift. The weighting
coefficients implemented are as follows: KU = 1000, KQ =
1000, and KH = 1. Fig. 10 shows the grid resulting phase
currents that represent the power terms also summarized in
Table II. Note the 81.37% reduction in the unbalance, the
80.82% reduction in the phase shift, and a nearly zero harmonic
distortion reduction (0.9%).The latter effect is also observable
in the clear third-order harmonic that keeps circulating through
the grid neutral.
Conversely, the third case assigns greater importance to the
harmonic distortion than to the other nonefficient phenomena.
In this regard, the weighting coefficients are implemented as
follows: KU = 1, KQ = 1, and KH = 1000. Fig. 11 shows the
grid currents measured after the compensation, and Table II
gathers the resulting power terms. Note the harmonic distortion
reduction (70.86%), and compare it to the harmonic reduction
experienced during the global compensation case initially in-
troduced (79.21%). The one during the selective compensation
is lower due to the power limitation introduced that prevents
ALFONSO-GIL et al.: OPTIMIZATION ALGORITHM FOR SELECTIVE COMPENSATION IN SAPF 3359

the SAPF from a better performance. The same phenomenon is

appreciated in Figs. 7 and 11.
Furthermore, some more cases have been analyzed, and
the resulting power compensations are compiled in Table II.
Among them, note the previously described case with KU =
1000, KQ = 1000, and KH = 1, in which the THDeI was
0.71. That THD is even higher than that existing before starting
the compensation (THDeI = 0.65 in Table I). This is because,
in that case, the SAPF compensates the grid fundamental cur-
rent components when canceling the load unbalance and phase
shift while the harmonic distortion is not modified. Therefore,
this reduction in the fundamental components implies the incre-
ment of the THDeI , according to (18). Also, note that, when
no preference is assigned to any of the nonefficient power terms
(case KU = KQ = KH = 1), the optimization cost index is
Fig. 12. Grid currents after compensation with the algorithm from [14].
exactly the sum of all the nonefficient phenomena. Thus, this is
the case that better profits the SAPF capability, and it is the one
recommended when there is no preference in the nonefficient
phenomena to cancel.
Finally, the presented algorithm is compared with that pro-
posed in [14], which also enables a SAPF to perform a selective
compensation of the nonefficient phenomena. In that work, the
authors introduce a priority-based compensation, where the first
preference is given to the harmonics. From there on, if the
SAPF is still powerful enough, the authors propose to reduce
the negative-sequence current, associated to unbalance, down to
5% of the full load current. Then, they propose to compensate
the reactive power until a power factor of 0.85 is achieved. Fi-
nally, if the SAPF still has some capacity left at this point, both
unbalance and reactive power are compensated with weightages
in the ratio of 4 : 1. This proposal mainly differs from the
Fig. 13. Grid currents after compensation with the LMI algorithm.
algorithm presented here in two aspects: First, the priority of
compensation in [14] is assigned on the basis of the considered and 13 show the results obtained with each algorithm. It can
gravity of the nonefficient effects, calculated from the SRF be seen how both approaches, as expected, almost completely
current decomposition, while in the cost index proposed here eliminate the harmonic currents. Furthermore, the unbalance is
in Section III-F, based on the IEEE Std. 1459 power terms, the also greatly reduced in both cases, although the LMI approach
weighting factors can be freely modified; second, [14] limits the gets a slightly better reduction. However, in terms of the reac-
algebraic sum of the current components, instead of the sum of tive power, the result with the [14] algorithm is clearly poorer,
squares as in (30). In this way, while this approach guarantees as can be seen from the current phase shift (cos = 0.99
that the SAPF currents do not exceed the filter limitations, in a versus cos = 0.90). The described conclusions are confirmed
general case, these currents will stay far from the real rms filter by the numerical quantification introduced in Table III. This
limit, wasting some of its compensation capacity. summarizes the nonefficient power terms according to the IEEE
To illustrate this point, a new simulation case comparing Standard 1459 shown for the noncompensated load and for both
both algorithms is developed (note that the load conditions cases. The difference between both algorithms behaviors is
of the previous cases cannot be replicated with the algorithm mainly due, as previously stated, to the different treatment of
from [14] because of the absence of a neutral wire in their the current limits. Indeed, the rms values of the filter currents
approach). In this case, the load contains a linear unbalanced are calculated for both cases: While the algorithm from [14]
three-phase load in parallel to a nonlinear balanced load. The gets Ia SAPF = 1.63 A, Ib SAPF = 1.60 A, and Ic SAPF =
former is constituted by resistances and windings (with values 1.15 A, the proposed LMI algorithm gets Ia SAPF = 2.46 A,
Laload = 230 mH, Lbload = 100 mH, and Rcload = 1000 ), Ib SAPF = 2.38 A, and Ic SAPF = 1.11 A, which are clearly
and the latter is constituted by a three-phase noncontrolled closer to the filter limits, therefore allowing a better overall
rectifier feeding an RC load through an inductive filter (with compensation.
values L = 5 mH, C = 2200 F, and R = 75 ). The SAPFs To conclude the comparison of these algorithms, it is im-
current limit is set in this case to 2.5 A. portant to highlight that, although the proposed LMI algorithm
In order to get a similar behavior in terms of the relative gets a better use of the filter capabilities in a general case, it
importance given to the different nonefficient phenomena from achieves this at the cost of a considerably higher computational
the LMI algorithm to the proposal in [14], the weighting coeffi- effort. Therefore, the algorithm from [14] remains a valid
cients are set to KU = 500, KQ = 1, and KH = 1000. Figs. 12 option for cases in which a very fast response of the selective
3360
TABLE III
P OWER T ERMS OF THE G RID C URRENTS
compensation algorithm could be required, i.e., situations in
which the load conditions changed very frequently.
VI. C ONCLUSION
When the power of all the nonefficient phenomena existing
in the grid is greater than the rated power of a SAPF introduced
in that grid to compensate them, it is desirable to have an
optimization algorithm that allows a selective compensation
without exceeding the SAPFs power. The proposed algorithm
is based on LMIs which allow to introduce quadratic constraints
in the calculation of the SAPF compensation currents to avoid
exceeding its rated power. In this way, it improves the perfor-
mance of SAPF during a selective compensation with regard to
other proposed works that do not consider current limitations
or are done inefficiently. In particular, this proposal has been
compared to that in [14], and with the same power constraints
and equivalent weighting coefficients, it achieves a 0.6% higher
reduction in SeN , a 7.1% higher reduction in SU 1 , and a 54.9%
higher reduction in Q+ 1.
The SAPF here presented implements a proportional feedfor-
ward current controller, which keeps the switching frequency
of the SAPF constant while offering a good dynamic response.
The results obtained both in simulation and in the laboratory
are illustrative of this point. In this regard, experimental results
demonstrate not only the proper functioning of the system
globally but also its capability to compensate selectively either
the unbalance, the reactive, or even the harmonic distortion of
the load currents.
R EFERENCES
[1] H. Akagi, E. Watanabe, and M. Aredes, More power to you (review of
Instantaneous Power Theory and Applications to Power Conditioning by
Akagi, H. et al.; 2007) [book review], IEEE Power Energy Mag., vol. 6,
no. 1, pp. 8081, Jan./Feb. 2008.
[2] S. Orts et al., Achieving maximum efficiency in three-phase systems with
a shunt active power compensator based on IEEE Std. 1459, IEEE Trans.
Power Del., vol. 23, no. 2, pp. 812822, Apr. 2008.
[3] M. Angulo, D. A. Ruiz-Caballero, J. Lago, M. L. Heldwein, and
S. A. Mussa, Active power filter control strategy with implicit closed-
loop current control and resonant controller, IEEE Trans. Ind. Electron.,
vol. 60, no. 7, pp. 27212730, Jul. 2013.
[4] F. Briz, P. Garcia, M. W. Degner, D. Diaz-Reigosa, and J. M. Guerrero,
Dynamic behavior of current controllers for selective harmonic compen-
sation in three-phase active power filters, IEEE Trans. Ind. Appl., vol. 49,
no. 3, pp. 14111420, May/Jun. 2013.
[5] J. He, Y. W. Li, and F. Blaabjerg, Flexible microgrid power quality
enhancement using adaptive hybrid voltage and current controller, IEEE
Trans. Ind. Electron., vol. 61, no. 6, pp. 27842794, Jun. 2014.
[6] S. Orts-Grau et al., Improved shunt active power compensator for IEEE
Standard 1459 compliance, IEEE Trans. Power Del., vol. 25, no. 4,
pp. 26922701, Oct. 2010.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 6, JUNE 2015
[7] Q.-N. Trinh and H.-H. Lee, An advanced current control strategy for
three-phase shunt active power filters, IEEE Trans. Ind. Electron.,
vol. 60, no. 12, pp. 54005410, Dec. 2013.
[8] H. Zhou et al., Selective harmonic compensation (SHC) PWM for grid-
interfacing high-power converters, IEEE Trans. Power Electron., vol. 29,
no. 3, pp. 11181127, Mar. 2014.
[9] J. C. Alfonso-Gil, J. J. Vague-Cardona, S. Orts-Grau, F. J. Gimeno-Sales,
and S. Segui-Chilet, Enhanced grid fundamental positive-sequence dig-
ital synchronization structure, IEEE Trans. Power Del., vol. 28, no. 1,
pp. 226234, Jan. 2013.
[10] F. Gonzalez-Espin, E. Figueres, and G. Garcera, An adaptive
synchronous-reference-frame phase-locked loop for power quality im-
provement in a polluted utility grid, IEEE Trans. Ind. Electron., vol. 59,
no. 6, pp. 27182731, Jun. 2012.
[11] A. Bhattacharya and C. Chakraborty, A shunt active power filter with
enhanced performance using ANN-based predictive and adaptive con-
trollers, IEEE Trans. Ind. Electron., vol. 58, no. 2, pp. 421428,
Feb. 2011.
[12] P. Acuna, L. Moran, M. Rivera, J. Dixon, and J. Rodriguez, Improved
active power filter performance for renewable power generation systems,
IEEE Trans. Power Electron., vol. 29, no. 2, pp. 687694, Feb. 2014.
[13] B. Singh, V. Verma, and J. Solanki, Neural network-based selective
compensation of current quality problems in distribution system, IEEE
Trans. Ind. Electron., vol. 54, no. 1, pp. 5360, Feb. 2007.
[14] B. Singh and V. Verma, Selective compensation of power-quality prob-
lems through active power filter by current decomposition, IEEE Trans.
Power Del., vol. 23, no. 2, pp. 792799, Apr. 2008.
[15] R. Gupta, Generalized frequency domain formulation of the switching
frequency for hysteresis current controlled VSI used for load compen-
sation, IEEE Trans. Power Electron., vol. 27, no. 5, pp. 25262535,
May 2012.
[16] S. Orts-Grau et al., Selective shunt active power compensator applied
in four-wire electrical systems based on IEEE Std. 1459, IEEE Trans.
Power Del., vol. 23, no. 4, pp. 25632574, Oct. 2008.
[17] IEEE Standard Definitions for the Measurement of Electric Power
Quantities Under Sinusoidal, Nonsinusoidal, Balanced, or Unbalanced
Conditions, IEEE Std 14592010 (Revision IEEE Std 14592000), 2010,
pp. 152.
[18] S. Orts-Grau et al., Selective compensation in four-wire electric systems
based on a new equivalent conductance approach, IEEE Trans. Ind.
Electron., vol. 56, no. 8, pp. 28622874, Aug. 2009.
[19] S. Orts-Grau, J. C. Alfonso-Gil, F. J. Gimeno-Sales, and S. Segui-chilet,
New resolution of the unbalance power according to Std. 1459, IEEE
Trans. Power Deliv., vol. 25, no. 1, pp. 341350, Jan. 2010.
[20] M. S. Lobo, L. Vandenberghe, S. Boyd, and H. Lebret, Applications
of second-order cone programming, Linear Algebra Appl., vol. 284,
no. 13, pp. 193228, Nov. 1998.
[21] S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan, Linear Matrix
Inequalities in System and Control Theory (Google eBook) 1994.
[22] J. Lofberg, YALMIP: A toolbox for modeling and optimization in
MATLAB, in Proc. IEEE Cat. No.04CH37508, 2004, pp. 284289.
[23] J. F. Sturm, Using SeDuMi 1.02, A Matlab toolbox for optimization over
symmetric cones, Optim. Methods Softw., vol. 11, no. 14, pp. 625653,
Jan. 1999.
[24] J. C. Alfonso-Gil, C. Ario, H. Beltrn, and E. Prez, Comparative study
of current controllers for shunt active power compensators used in smart
grid applications, in Proc. ICREPQ, Bilbao, Spain, 2013, pp. 16.
[25] Y. Ogata, Katsushiko, and Yang, Modern Control Engineering.
Englewood Cliffs, NJ, USA: Prentice-Hall, 1970.
Jos Carlos Alfonso-Gil received the M.E.
degree in automation and industrial electronics
engineering from the Universidad Politcnica de
Valencia, Spain, in 2004 and the Ph.D. degree
in electronics engineering in 2010.
Since 2007, he has been an Assistant Profes-
sor in the Electrical Engineering Area at the Uni-
versitat Jaume I, Castellon de la Plana, Spain.
His major fields of interest are active power
filters, control of power converters, renewable
energy systems, and microgrids.
ALFONSO-GIL et al.: OPTIMIZATION ALGORITHM FOR SELECTIVE COMPENSATION IN SAPF
Emilio Prez received the M.Sc. degree in in-
dustrial engineering from the Universitat Jaume
I (UJI), Castellon de la Plana, Spain, in 2002
and the Ph.D. degree in control engineering
from the Universidad Politcnica de Valencia
(UPV), Spain, in 2011.
From 2002 to 2006, he worked with the Pre-
dictive Control and Heuristic Optimization group
at UPV. Since 2006, he has been with the
Electrical Engineering Area at UJI, where he
is currently an Assistant Professor. His current
research interests include model predictive control, convex optimization,
control of PV plants with energy storage, and active power filters.
C. Ario received the M.Sc. degree in industrial
engineering and the Ph.D. degree in control
engineering from the Universidad Politcnica de
Valencia, Spain, in 2003 and 2008, respectively.
Since 2006, he has been with the Industrial
Systems Engineering and Design Department,
Universitat Jaume I, Castellon de la Plana,
Spain. His current research interests include
fuzzy control theory, predictive control, intelli-
gent control, and microgrids.
3361
Hector Beltran received the M.Sc. degree in in-
dustrial engineering from the Universitat Jaume
I (UJI), Castellon de la Plana, Spain, in 2004
and the Ph.D. degree in electrical engineer-
ing from the Technical University of Catalonia
(UPC), Spain, in 2011.
In 2003, he worked at the European Cen-
tre for Nuclear Research (CERN), Geneva,
Switzerland. From 2004 to 2006, he worked
at the Institute of Energy Technology, Valencia,
Spain. Since 2006, he has been an Assistant
Professor in the Electrical Engineering Area at UJI. His current research
interests include massive PV integration into the electric power system,
energy storage systems, and microgrids.