Expert Systems with Applications 42 (2015) 35623570

Contents lists available at ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier.com/locate/eswa

Novel expert system for dening power quality compensators

Eduardo Verri Liberado a,, Fernando Pinhabel Marafo b, Marcelo Godoy Simes c,
Wesley Angelino de Souza a, Jos Antenor Pomilio a
a
University of Campinas (UNICAMP), School of Electrical and Computer Engineering (FEEC), Department of Systems and Energy (DSE), Av. Albert Einstein,
400, 13083-852 Campinas, SP, Brazil
b
Univ Estadual Paulista (UNESP), Campus of Sorocaba, Group of Automation and Integrated Systems (GASI), Av. Trs de Maro, 511, 18087-180 Sorocaba, SP, Brazil
c
Colorado School of Mines, Department of Electrical Engineering and Computer Science, 80401 Golden, CO, United States

a r t i c l e

i n f o

Article history:
Available online 27 December 2014
Keywords:
Compensation
Decision-maker
Expert system
Harmonic distortion
Load unbalance
Power factor correction
Power quality

a b s t r a c t
In order to ensure good power quality for modern power systems and/or industrial installations, power
conditioning devices have been extensively applied. However, the data analysis for the installation of a
determined compensator mainly considers a particular power quality index or disturbance and it is
usually based on human expertise. Therefore, this paper proposes a novel expert system that automatically suggests the most appropriate and cost-effective solution for compensating reactive, harmonic and
unbalanced current through a careful analysis of several power quality indices and some grid characteristics. Such an expert system is an important tool in real-world applications, where there is a complex
scenario in choosing, designing and applying power quality compensators in modern power grids. Since
there are no strict boundaries for voltage and current non idealities at distribution level or clear
correlation between them and possible solutions, a fuzzy decision-maker was developed to deal with
such uncertainties and to embed human expertise in the system. The approach is based on analyzing data
from a given time window and providing off-line recommendations for the design and installation of
proper compensators. Therefore, the application of the proposed expert system may result in enhanced
and faster projects when compared to the traditional design methods for power conditioning. A
computational study consisting on applying the suggested compensators for a 5-node network and
different load congurations shows the effectiveness of the proposed expert system.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
In terms of industrial applications, power conditioning devices
may ensure good power quality (PQ) at a certain point of common
coupling (PCC). This consists in avoiding that particular load characteristic, such as unbalanced connection, reactive power and non
linearities (which cause harmonic distortion) affect the grid. Consequently, the installation of a compensator may increase the network efciency, reduce costs and avoid consumers penalizations
by low power factor (Akagi, Watanabe, & Aredes, 2007; Bollen,
2003; Singh, Al-Haddad, & Chandra, 1999).
Even though many traditional compensators (such as capacitor
banks for power factor correction) are still used and might be costeffective, there are an increasing number of restrictions for their
application, mainly due to high levels of voltage distortion in the
Corresponding author. Tel.: +55 15 98118 3886.
E-mail addresses: evldoc@dsce.fee.unicamp.br (E.V. Liberado), fmarafao@
sorocaba.unesp.br (F.P. Marafo), msimoes@mines.edu (M.G. Simes), wesley@fc.
unesp.br (W.A.d. Souza), antenor@fee.unicamp.br (J.A. Pomilio).
http://dx.doi.org/10.1016/j.eswa.2014.12.032
0957-4174/ 2014 Elsevier Ltd. All rights reserved.

grid (Bisanovic, Hajro, & Samardzic, 2014; Jintakosonwit,

Srianthumrong, & Jintagosonwit, 2007). In this case, the compensators may cause more problems than tangible benets (Currence,
Plizga, & Nelson, 1995; Obulesu, Reddy, & Kusumalatha, 2014;
Phipps, Nelson, & Sen, 1994).
Therefore, during the design stage of a compensation system, it
is important to consider several PQ indices, observing its recommended limits (when available), in order to determine the feasibility of such solution.
Nevertheless, the main problems to proceed in this way are: the
absence of PQ indices in standard power meters (usual frontier
equipment between industries and utilities) and/or contradictory
results for a PQ index when calculated by different power meters
(Galvo, Belchior, Silveira, & Ribeiro, 2014); the lack of strict
boundaries for all voltage and current non idealities at distribution
level and specially, the absence of clear relationship among them
and possible solutions; and nally, the lack of expertise on the
application of power conditioners by young practical engineers.
Regarding the PQ monitoring, the current literature has leading
to new trends in electrical grids, mostly related to the concept of

3563

E.V. Liberado et al. / Expert Systems with Applications 42 (2015) 35623570

smart grids and distributed generation systems. In this scenario,

the proposed power meters have been named smart meters
(Marafo, Souza, Liberado, Silva, & Paredes, 2013) and they are able
to inform PQ indices and relevant information to consumers and
utilities. Moreover, smart metering has been applied to PQ monitoring and assessment of grids with distributed renewable sources
(Chompoo-Inwai & Mungkornassawakul, 2013; Golovanov,
Lazaroiu, & Porumb, 2013; Su et al., 2013; Valle et al., 2013;
Zhang, Yan, Yang, Bao, & Sun, 2013; Hashemi & Aghamohammad,
2013; Yang et al., 2013) and PQ event diagnosis (Faisal,
Mohamed, Shareef, & Hussain, 2011; Gunal, Gerek, Ece, &
Edizkan, 2009; Kriukov, Grigoras, Scarlatache, Ivanov, & Vicol,
2014; Salem, Mohamed, & Samad, 2010; Wang & Tseng, 2011). In
such applications, some PQ indices are calculated, the voltage prole is estimated and the data is properly processed, but the proposed systems just provide information that must be analyzed by
end-user in order to make a decision about the need of power conditioning for ensuring pre-dened PQ levels.
In terms of expert systems related to PQ compensators, some
recent applications are optimal planning of passive lters
(Bhattacharya & Goswami, 2009; Chang, 2010; Chang, Low, &
Hung, 2009; Low, Chang, & Hung, 2009) and control of FACTS (Flexible AC Transmission Systems) (Banerjee, Mukherjee, & Ghoshal,
2014; Suslov, Solonina, & Smirnov, 2014) and inverters (Cheng,
2011; Jegathesan & Jerome, 2011; Ray, Chatterjee, & Goswami,
2010; Tutkun, 2010). However, those authors are focused on a particular sort of compensators and, predominantly, considering the
PQ indices related to harmonic distortion.
On the other hand, expert systems have been extensively
applied in industrial and power systems, power electronics and
related areas, as well as revised by Hassan, Moghavvemi, Almurib,
and Steinmayer (2013) and Sahin, Tolun, and Hassanpour (2012).
Additionally, a related application of expert systems that should
be highlighted is the denition of new PQ factors based on the ones
traditionally used to quantify the PQ (Arghandeh, von Meier, &
Broadwater, 2014; Morsi & El-Hawary, 2008a,b).
Therefore, this work proposes an expert system that calculates
relevant PQ indices in order to characterize the most signicant
ones, analyses those PQ indices and, as a result for the end-user,
denes a proper PQ compensator to improve the PQ on a particular
point of common coupling (PCC).
On the proposed approach, the decision process is found on a
Fuzzy Decision-Maker (DM) (Bellman & Zadeh, 1970; Caia,
Huangc, Lina, Niea, & Tana, 2009; Piltan, Mehmanchi, & Ghaderi,
2012), which is typically used when the expected output is a list
of linguistic alternatives (decisions) and the defuzzication is not
applicable or necessary. Thus, this fuzzy decision process is responsible for correlating PQ indices and grid characteristics through a
Fuzzy Rule-Based System (FRBS) towards representing the human
expertise in compensation technologies. The decision-maker is
based on off-line analysis of previously measured data, within a
given time window.
The next sections describe the adopted PQ indices and grid
characteristics, the proposed expert system, its application (in simulation) on a grid with different load congurations and the results
of installing the suggested compensators for each load condition.

Institute of Electrical and Electronics Engineers, 2010), which

deal with the most important PQ disturbances and might offer useful information to determine proper compensation systems.
Moreover, there are some grid characteristics that might be
considered to determine a suitable compensator, such as the short
circuit level (SCl) and the dynamic behavior of the fundamental
positive sequence reactive power (Q1+) (IEEE Institute of
Electrical and Electronics Engineers, 2010). These characteristics
and the selected PQ indices are described in the following.
2.1. Total harmonic distortion (THD) and total demand distortion
(TDD)
The THD index (i.e., the ratio between the harmonic components and the fundamental frequency component of voltages or
currents) is used to quantify the harmonic distortion of voltages
(THDv), in order to determine if voltage distortions could damage,
e.g., the installation of a capacitor bank.
The limits adopted for THDv follows the IEEE 519-1992 (IEEE
Institute of Electrical and Electronics Engineers, 1993), which recommends that for voltage levels under 69 kV the THDv should not
be bigger than 5%.
In order to determine if harmonic current mitigation is necessary, the total demand distortion (TDDi) is adopted (IEEE
Institute of Electrical and Electronics Engineers, 1993), which is
the ratio between the harmonic currents and the maximum
demand load current.
In fact, the main goal of the harmonic current mitigation is to
avoid that the harmonic currents affect the supply voltages
through the supply/line impedance. Thus, an important grid characteristic to be considered during the PQ analysis and is often
used by IEEE (1993) to determine the TDDi limits is the short-circuit level (SCl) of the PCC, which is the ratio between the short circuit current (or short circuit apparent power) and the maximum
load current/apparent power demand.
In other words, in a grid with high SCl (the so called strong
grid) the current demand is low and the current distortion would
not signicantly affect the grid, thus in this case the TDDi limits are
greater than in a PCC with a small SCl (which is called weak grid),
which is shown in Table 1, extracted from IEEE (1993) for voltages
under 69 kV.
2.2. Unbalance factor (K)
The unbalance factors consider fundamental sequence components and may be useful to quantify PCC voltage and currents
asymmetries (IEEE Institute of Electrical and Electronics
Engineers, 2010). Eq. (1) presents the negative sequence (K) and
the zero sequence (K0) unbalance factors for a three phase system:

KX 

X 1
X0
KX 0 1 :100
 100
X1
X1

on which X1 is the RMS value of the fundamental positive, negative

or zero sequence of voltage or currents signals.
The unbalance factors for the voltages (KV and KV0) are limited
in 2% for voltage levels greater than 13.8 kV. However, for voltage

2. Power quality indices and grid characteristics

In order to quantify the power quality at a given PCC, several PQ
indices may be calculated and analyzed. Some of them have
recommended limits that may be used for utilities to penalize consumers (or loads) that do not respect the predened bounds.
For the development of the proposed expert system, the
authors chose to use indices extracted from IEEE recommendations
(IEEE Institute of Electrical and Electronics Engineers, 1993; IEEE

Table 1
Adopted limits for total demand distortion of currents.
SC1

TDDi (%)

<20
2050
50100
1001000
>1000

5.0
8.0
12.0
15.0
20.0

3564

E.V. Liberado et al. / Expert Systems with Applications 42 (2015) 35623570

levels under 13.8 kV or even for the unbalance factors calculated

for the currents (KI and KI0) there are not recommended limits.
Thus, the authors adopted that both negative and zero sequence
unbalance factors calculated for voltages may be limited to 5%,
while the ones calculated for currents may be limited to 15%. Such
limits were established by means of intensive communication with
specialists from industry, utilities and academy.

PCC

A
V

Grid

2.3. Power factor (PF)

The PF, dened as the ratio between the active power and the
apparent power, is one of the most important indices for PQ evaluation. It is commonly used by the utilities to penalize the consumers for excessive reactive power demand, and consequently, it is a
real stimulus to install compensation systems.
However, as PF is reduced not only by the presence of reactive
power but also due to other PQ disturbances, such as harmonic distortion and load unbalance (IEEE Institute of Electrical and
Electronics Engineers, 2010), it is not recommended that it would
be the unique parameter used for PQ evaluation.
Considering different international standards, the limits for PF
can differ considerably. So, in this application, it was adopted that
PF at a certain PCC should be higher than 0.95.

2.4. Reactive power dynamics

In addition to the SCl, other important characteristic to be considered on the PCC analysis is the behavior of the reactive power
over the time. As the load changes along a certain period of time
(and obviously the reactive power too), it is important that the
reactive power compensators follow large load variations in order
to avoid over/under-compensation and other detrimental issues.
Thus, the authors assume that if the reactive power (Q1+) at a
given PCC has a variation of 30% or more during the measurement
period, the compensators should be adjustable in terms of reactive
power. Such boundaries were also dened based on specialists
recommendations.

Load

Voltage, current,
power components
and PQ indices

limits for
PQ indices
and Q1+

% of measurement
time exceeding the
limit

SCl at PCC

FRBS

Expert system

Compensation
solution

Measurement
during a
determined
time interval

Data
processing

Analysis

Fig. 1. Functional diagram of the expert system for power quality improvement
performance.

If the PQ index is out of boundaries for a small percentage of the

measuring time, it may indicate that it is a temporary problem
or it is not economically interesting to be compensated.
Otherwise, it may indicate that it is a critical problem at that
PCC and it is very recommendable to be compensated.
Besides, the association of processed information to fuzzy membership functions and the decision-maker becomes very intuitive.
Next section describes the developed off-line DM, which proposes
a particular compensation solution to a certain PCC, based on the
analysis of pre-processed PQ data.

3. The proposed expert system

The proposed expert system is described according to Fig. 1.
Three-phase voltages and currents are measured at a certain PCC
during a given time interval, which might be large enough to
describe all possible load characteristics and PQ disturbances at
the PCC. It means that such time interval depends on the load
dynamics and it may be dened considering information provided
by the consumer.
During the measurement interval, power components and PQ
indices are calculated and stored using a specic sampling rate,
which depends on the measurement total time and storage capacity of the power meter. To calculate the TDDi of each phase, the rms
values of the harmonic currents are stored and post-processed considering the maximum current demand at the measuring point.
Therefore, the expert system is based on off-line analysis of previously stored data. Thus, stored samples of the indices described
in Section 2 are processed in a computer in order to identify the
percentage of time, on which every index exceed the pre-dened
limits. In case of THDv and TDDi, only the maximum value over
the three phases is considered. In case of the unbalance factors, just
the highest value between negative and zero sequence factors is
considered. This approach allows dening a degree of relevance
or degree of compensation feasibility for each PQ disturbance
identied at the PCC:

4. A fuzzy rule-based expert system

The DM for the proposed expert system consists in a FRBS with
seven inputs and two output variables. The inputs are the percentage of measurement time in which the ve PQ indices (PF, THDv,
TDDi, KV, KI) and Q1+ variation (dQ) are out of their pre-dened limits and the value of SCl at the PCC. Fig. 2 presents the membership
functions (MF) of the inputs and outputs.
The input MFs were dened using trapezoidal shape in order to
express that the need for compensation of each PQ disturbance is
linearly related to their incidence during the measuring time. In
this sense, for those inputs related to PQ indices, two linguistic
terms were dened: EST (exceeded the limit for a short time),
which means that the PQ disturbance is occasional and/or its compensation is negligible; and ELT (exceeded the limit for a long
time), which means that the PQ problem is persistent and should
be compensated. For SCl, the linguistic terms are simply low (L)
and high (H). The range of each MF was dened according to specialists expertise in compensation systems. It would be possible to
dene more linguistic terms, such as exceeded the limit for a very
long time or exceeded the limit for a very short time, but the
addition of more linguistic terms would not improve, signicantly,
the DM response.

E.V. Liberado et al. / Expert Systems with Applications 42 (2015) 35623570

(a) MF for PF

(b) MF for THDv

(c) MF for TDDi

(d) MF for KV

(e) MF for KI

(f) MF for dQ

(h) MF for reactives

(i) MF for harmonics

(g) MF for SCl

3565

Fig. 2. Membership functions.

Although the inputs have only two linguistic terms, suggesting

that a Boolean approach (in which the input values must be
inserted in only one of the sets, and outputs would be determined
by a truth table) could be also used, the fuzzy approach offers some
advantages regarding to a Boolean method, such as:
The limits of Boolean sets should be sharply dened, but there
are not recommended boundaries available for the inputs of
the proposed DM. So, by using fuzzy sets, the uncertainty of
boundaries denition can be considered in the model.
The use of fuzzy sets results that a determined number of elements can have a non-zero membership degree in two or more
sets and this means that some input values might have two or
more possible output values, i.e., suggested compensation solutions. In real applications this would help the user in the direction of a nal decision by considering additional criteria (e.g.,
cost of installation, complexity of the compensator design,
effectiveness of the compensation, etc.). This condition is not
feasible when using a Boolean approach.
The two output variables dene a set of compensation systems
for the analyzed PCC. However, it is important to mention that the
design and control of the recommended compensators are subsequent steps, which are not discussed in this paper. The variable
called reactives represents linguistic terms for compensators aimed
to reactive power minimization (xed capacitor banks, thyristorswitched capacitors TSC, active lters) and load unbalance (static
var compensator SVC) compensation. The linguistic terms from
the variable harmonics are the proposed compensators for harmonic distortion (passive lters and active lters).
Moreover, both output variables have the linguistic term donot-compensate, which represents that compensation is not necessary or may not be possible using the considered compensators.
The output MFs were dened as triangular functions just to obtain
a linear relationship with the degrees of membership. Thus, if the
fuzzy inference process results more than one compensation solution, the expert system may suggest the one with maximum
degree of membership.

The use of two distinct output variables allows the DM to suggest cooperative compensation strategies (Cheng, Chen, Lee, & Kuo,
2009; Paredes, Costabeber, & Tenti, 2011a; Rahmani, Hamadi,
Al-Haddad, & Dessaint, 2014), on which a combination of
compensators solves both reactive and harmonic disturbances.
This approach improves the compensation results, avoids detrimental interaction among compensators and offers cost-effective
solutions, instead of just suggesting active lters to solve all PQ
problems at a given PCC.
Moreover, it is important to mention that compensation solutions proposed by the expert system are aimed to minimize PQ disturbances correlated to the currents. Therefore, the input variables
associated to the voltages indices are used to restrict the use of
some compensators when the voltages are unbalanced and/or
distorted.
The rules dened in the FRBS for the DM are listed in Table 2, on
which dashed elds indicate that the input variable is a do not
care input for that compensation solution (e.g., SVC is suggested
regardless Q1+ variation due to their built-in exibility to adjust
their parameters to follow or not, if there is not Q1+ variation).
Each line of this table might be read as the following example
(rule 7):
If FP is ELT and DHTi is ELT and KV is EST and KI is EST and dQ
is EST and SCl is L then reactives is xed-capacitor-bank, harmonics
is passive-lter.
Such rules relate human expertise, PQ limits in terms of standard recommendations, as well as the percentage of time on which
the limits are violated to possible solutions. Capacitor banks are
suggested to compensate reactive power and increase power factor
but only if THDv is under the accepted limits (at least for most part
of time) or in presence of harmonic distortion compensators.
Moreover, if there is Q1+ variation, a thyristor-switched capacitor
(TSC) is chosen instead of xed capacitors. In the presence of
unbalanced load, the expert system suggests SVC, which also compensates reactive power, but produces harmonic distortion due to
reactors switching (Gyugy, Otto, & Putman, 1978). So, whenever
the expert system suggests SVC, its intrinsic passive lters should
be applied in order to compensate its inherent current distortion.

3566

E.V. Liberado et al. / Expert Systems with Applications 42 (2015) 35623570

Table 2
Rule based of the proposed expert system.

1
2
3
4
5
6
7
8
9
10
11
12
13

PF

THDv

TDDi

KV

KI

dQ

SCl

reactives

harmonics

EST
ELT
ELT
ELT
ELT
EST
ELT
ELT
ELT
EST
ELT
ELT
ELT

EST
EST
EST
ELT
EST

EST
EST
EST
EST
EST
ELT
ELT
ELT
ELT
ELT
ELT
ELT
ELT

EST
EST
EST
EST

EST
EST
EST

EST
EST
EST

EST
EST
EST
EST
ELT
EST
EST
EST
ELT
EST
EST
EST
ELT

EST
ELT

EST
EST
ELT

EST
ELT

L
L
L
L
H
H
H
H

do-not-compensate
xed-capacitor-bank
TSC
active lter
SVC
do-not-compensate
xed-capacitor-bank
TSC
SVC
do-not-compensate
xed-capacitor-bank
TSC
SVC

do-not-compensate
do-not-compensate
do-not-compensate
do-not-compensate
do-not-compensate
passive-lter
passive-lter
passive-lter
passive-lter
active-lter
active-lter
active-lter
active-lter

Regarding to harmonic current compensation at the PCC, if SCl

is low (high line impedance), the efciency of passive lters is very
good, so that it should be the more suitable suggestion. However, if
SCl is high (low line impedance), active power lter is a most
appropriate solution. Other condition on which active lter would
be suggested is, e.g., in case of reactive compensation under distorted grid voltages, when the installation of xed capacitors
may cause harmonic resonances and it is not recommended.
5. Simulation results
In order to evaluate the proposed expert system, a three phase
5-node grid was simulated on PSCAD considering 5 different load
congurations. Fig. 3 depicts the simulated power circuit.
The loads are connected to node E (low voltage level), while the
PCC is assumed to be at node D (medium voltage level), where PQ
compensators are commonly installed. The PCC short-circuit current is 1584.98 \69A. Thus, next two sub-sections show the
results of applying the expert system using PCC measurements,
for each load conguration and then, the results of applying the
solutions proposed by the DM in each case.
5.1. Application of the expert system
Table 3 describes the load congurations and the time interval
on which each load was connected to node E. Case 1 and case 6
have same load congurations, but in 6 it was added 6% of 5th harmonic in the voltages at node C. For each case, total simulation and
measurement time were 8.0 s.
Table 4 shows samples of stored power quality data from each
load condition and the correspondent time instant when they were
calculated. The time rate for storing the calculated power components and PQ indices was dened on a 60 Hz basis, it means, one
sample per fundamental period. Thus, the stored data obtained
from the measurement device were processed and analyzed by
the DM using Matlab.
Table 5 depicts the input variables for the DM, the maximum
RMS value of PCC currents (used to calculate the SCl and the TDDi
samples shown on Table 4) and the compensation solutions
suggested by the expert system. In addition, Table 6 shows the

activated rules and which one was selected by the DM, as the most
proper solution for each case.
Case 1 was a typical application of PF correction, where the load
is balanced, constant and free of harmonic distortion. Thus, the corresponding rule from FRBS for this case was number 2, resulting
xed-capacitor-bank for reactives and do-not-compensate for
harmonics.
Nevertheless, in case 6 (same load of case 1, but assuming distorted PCC voltages) the harmonic distortion identied at PCC voltages was higher than the adopted limit and, although it did not
affect too much the currents (TDDi was 1.859%), it could affect signicantly the compensation if a capacitor bank would be applied.
Thus, the suggestion of the expert system is active-lter for reactives and do-not-compensate for harmonics (rule 4).
In the next section the result of using just a capacitive bank for
case 6 will be discussed to conrm that it would not be a good
choice.
In case 2, the existence of reactive power and distorted currents
contribute for a low PF during all the measurement interval. Consequently, the PCC voltages became distorted, but THDv is still
under the recommended limit. In this case, the short circuit level
has non-zero membership degrees for both MFs of SCl. Consequently, rules 7 and 11 from FRBS are activated (xed-capacitorbank for reactives, and passive-lter or active-lter for harmonics).
However, one can observe in Table 6 that as the degree of membership of passive-lter is bigger than the degree of membership of
active-lter, DM suggested rule 7.
In case 3, the load varies during the measurement time, resulting in four different load proles, as shown in Table 4. The average
Q1+ is 344.2 kVAr and considering the values of Q1+ from Table 4,
this power component is out of the adopted boundaries (30%
deviation) for 74% of total measurement time, as indicated in
Table 5. Moreover, the current distortion exceeds the pre-dened
limit during 42% of measurement time, which corresponds to the
time when the rectier is turned on. Thus, by considering the reactive power variation and the harmonic distortion during a signicant time interval, at the intermediary value of short circuit level,
the activated rules in the FRBS were 8 and 12. In accordance to
the membership degrees illustrated in Table 6, the decision-maker
suggests rule 8 as the outcome of the decision making process.

Fig. 3. 5-node power circuit simulated at PSCAD.

3567

E.V. Liberado et al. / Expert Systems with Applications 42 (2015) 35623570

Table 3
Load conguration for each simulation.
Case

Time interval connected to grid

[s]

Load

1, 6

08.0

Three-phase delta-connected balanced RL load: R = 150 mO and L = 300 lH between phases

08.0

Three-phase thyristor rectier with DC RL load: RDC = 200 mO and LDC = 20 mH

Three-phase delta-connected balanced RL load: R = 211 mO and L = 420 lH between phases

02.0; 6.08.0
2.07.34
2.666.0

Three-phase delta-connected balanced RL load: R = 620 mO and L = 1.2 mH between phases

Three-phase delta-connected balanced RL load: R = 211 mO and L = 420 lH between phases
Three-phase thyristor rectier with DC RL load: RDC = 200 mO and LDC = 20 mH

08.0

Three-phase delta-connected balanced RL load: R = 300 mO and L = 600 lH between phases

Three-phase wye-grounded unbalanced RL load: RA = 125 mO; LA = 230 lH; RB = 250 mO; LB = 450 lH; RC = 63 mO;
LC = 120 lH

08.0

Three-phase thyristor rectier with DC RL load: RDC = 200 mO and LDC = 20 mH

Three-phase wye-grounded unbalanced RL load: RA = 125 mO; LA = 230 lH; RB = 250 mO; LB = 450 lH; RC = 63 mO;
LC = 120 lH

Table 4
Data samples stored at each load conguration.
Case

t [s]

PF

THDvA (%)

THDvB (%)

THDvC (%)

TDDiA (%)

TDDiB (%)

TDDiC (%)

KI (%)

KI0 (%)

KV (%)

KV0 (%)

S [kVA]

P [kW]

Q1+ [kVAr]

1
2
3

8.0
8.0
2.5
5.0
7.0
8.0
8.0
8.0
8.0

0.7735
0.7923
0.7973
0.8041
0.7941
0.7844
0.7706
0.7911
0.7722

0.006
3.176
0.08
3.101
0.004
0.011
0.006
3.212
5.986

0.012
3.176
0.02
3.101
0.008
0.005
0.012
3.287
5.986

0.006
3.176
0.01
3.101
0.005
0.005
0.006
3.217
5.986

0.016
11.574
0.015
11.507
0.015
0.022
0.015
12.797
1.859

0.011
11.574
0.009
11.507
0.022
0.096
0.015
17.872
1.853

0.016
11.574
0.016
11.507
0.031
0.083
0.015
13.814
1.859

0.005
0.003
0.007
0.004
0.003
0.013
17.002
17.511
0.005

0.005
0.003
0.007
0.004
0.004
0.014
0.006
0.005
0.005

0.004
0.002
0.007
0.003
0.003
0.005
0.386
0.399
0.004

0.004
0.002
0.007
0.003
0.003
0.004
0.004
0.004
0.004

730.3
878.7
509.6
843.6
677.8
189.1
722.6
729.3
731.7

564.8
696.2
406.3
678.4
538.2
148.3
556.9
577.0
565.0

462.9
524.8
307.7
490.4
412.0
117.2
444.1
413.7
462.9

4
5
6

Table 5
FRBS input and output variables and maximum RMS current in each case.
Case

PF
(%)

THDv
(%)

TDDi
(%)

KV
(%)

KI
(%)

100

100

100

dQ
(%)

Maximum RMS current

[A]

SCl

Activated rules
(Table 2)

reactives

harmonics

36.0

44.02

do-notcompensate
passive-lter

8
12
5

xed-capacitorbank
xed-capacitorbank
xed-capacitorbank
TSC
TSC
SVC

43.4

36.52

9
13
4

SVC
SVC
active lter

11
3

100

42

74

41.6

38.10

100

100

39.7

39.92

100

100

100

39.9

39.75

100

100

36.0

44.02

In case 4, the unbalanced load is characterized by means of the

negative sequence factor KI (KI0 was almost zero due to the deltagrounded wye conguration of Transformer 2). Thus, the DM indicates that the best solution should be the application of a SVC to
compensate reactive power and load unbalance (rule 5).
In case 5, the FRBS proposes a SVC for reactive power and unbalanced currents and, passive-lter or active-lter (rules 9 and 13) for
mitigating harmonic components generated not only by SVC, but
also by the operation of the load (rectier). However, as the membership degree of passive lter was bigger than active-lters, the
suggested solution was rule 9.
5.2. Compensators implementation
This section discusses the application of the solutions proposed
by the expert system, considering each case mentioned on previous section. The simulation time was 8.0 s in each case. Table 7

active-lter
passive-lter
active-lter
do-notcompensate
passive-lter
active lter
do-notcompensate

illustrates some samples of power components and PQ indices calculated at the same instant than the ones presented on Table 4;
however, assuming now the operation of the suggested compensators for each case.
In cases 15, the passive compensators were designed according to the methodologies proposed in Liberado, Souza, Pomilio,
Paredes, and Marafo (2013). PQ indices were maintained under
their recommended limits, Q1+ was considerably reduced and the
association of reactive/unbalance compensators with harmonic
compensators resulted in almost unitary PF.
In case 6, the suggested active lter was connected to node E
(low voltage level, Fig. 3) to operate as a current source injecting
the corresponding reactive component of load current (Paredes,
Silva, Brando, & Marafo, 2011b). According to Table 7, the compensation of reactive power at PCC was very effective, power factor
was closed to unity and harmonic distortion levels were not
altered and did not affect the compensation.

3568

E.V. Liberado et al. / Expert Systems with Applications 42 (2015) 35623570

Table 6
Graphical representation of the fuzzy rules for each case.

Case

Rules

PF (%)

THDv (%)

TDDi (%)

KV (%)

KI (%)

dQ (%)

SCl

reactives

harmonics

Solution
rule 2

rule 7

11
8

rule 8

12

rule 5

rule 9

13

rule 4

Table 7
Data samples stored for each case considering the compensators turned on.
Case

t [s]

PF

THDvA (%)

THDvB (%)

THDvC (%)

TDDiA (%)

TDDiB (%)

TDDiC (%)

KI (%)

KI0 (%)

KV (%)

KV0 (%)

S [kVA]

P [kW]

Q1+ [kVAr]

1
2
3

8.0
8.0
2.5
5.0
7.0
8.0
8.0
8.0
8.0
8.0

0.9999
0.9991
0.9999
0.9989
0.9999
0.9999
0.9976
0.9968
0.9974
0.9475

0.011
1.013
0.066
1.154
0.017
0.059
0.219
0.813
5.988
8.833

0.015
1.013
0.022
1.154
0.020
0.048
0.218
0.738
6.003
8.839

0.011
1.013
0.029
1.154
0.017
0.050
0.181
0.776
6.014
8.829

0.027
3.893
0.029
4.129
0.011
0.075
1.881
5.058
1.600
32.719

0.067
3.897
0.010
4.157
0.057
0.020
3.246
5.581
1.587
32.689

0.089
3.975
0.012
4.178
0.073
0.026
2.793
5.871
1.599
32.735

0.003
0.002
0.003
0.002
0.002
0.007
0.386
1.277
0.011
0.003

0.003
0.002
0.004
0.002
0.002
0.007
0.007
0.007
0.005
0.003

0.004
0.003
0.002
0.003
0.003
0.007
0.006
0.024
0.007
0.009

0.004
0.004
0.001
0.003
0.004
0.008
0.004
0.004
0.004
0.004

580.3
728.8
413.7
708.0
551.4
149.4
576.7
606.5
634.8
612.9

580.3
728.1
413.7
707.2
551.4
149.4
575.3
604.6
633.1
580.7

2.6
11.0
9.0
11.2
2.2
6.1
36.6
33.0
30.8
1.0

4
5
6
6

Just to illustrate the problems of applying traditional power factor correction on a PCC with distorted voltages, case 6 was simulated with the capacitor bank in the same way to case 1, instead
of using the suggested active power lter. The results are shown
in Table 7 and indicated as case 6. In such case, even though the
compensation of Q1+ was effective, the presence of capacitors produces a resonant effect that increases THDv and TDDi. Consequently, the compensation give rise to more PQ problems than
solutions (neither PF reached their minimum limit, which was
0.95). In fact, this case should be compensated for shunt active
lters or even some more complex and costly solutions, such as a
combination of series compensator for the voltage harmonic
distortion and passive shunt compensators (Corasaniti, Barbieri,
Arnera, & Valla, 2009; Jintakosonwit et al., 2007).
6. Conclusions
Currently, the development and use of power quality compensators is essential for the operation of most modern power grids.
If correctly chosen, designed and controlled, such power conditioning devices are able to minimize power losses, as well as several
voltage and current disturbances over the grid, maximizing the
grids efciency or simply keeping it under stable and standardized
operation. Depending on their physical topology, they may be
applied to ensure good voltage quality for sensible loads or to minimize the impact of disturbing load behavior over the grid.
In such context, this paper shows a novel expert system able to
decide the best compensation solutions for power quality disturbances caused by load. No parallel of the proposed system was
found in related literature.
The theoretical contributions are found on the fact that, instead
of focusing on PQ indices related to one particular PQ disturbance
and/or compensator, such off-line system is based on the

simultaneous analysis of several power quality indicators and grid

characteristics, under a given time window. Moreover, the analysis
of the grid information is made by the expert system in order to
suggest, automatically, appropriate solutions for different supply
and load conditions.
Considering the absence of particular boundaries for all voltage
and current disturbances and the lack of clear relationship among
them and possible solutions, the expert system was found on specialists expertise, by means of a fuzzy rule-based system, and
organized as a PQ compensation oriented fuzzy decision-maker,
as presented on Fig. 2 and Table 2 and discussed on Section 4.
Human experience was included mainly for dening the borders
of some power quality indices (those not dened on specic standards) and the period of time they could gure out of such limits,
as well as for correlating such uncertain constrains with feasible
and cost-effective compensation alternatives, which addresses
the nal decision to the user.
Some practical contributions that may be highlighted are: the
analysis of the time period on which the PQ indices are out of their
limits help to dene a degree of relevance for the membership
functions of the FRBS. Based on such information, the decisionmaker is able to evaluate if a particular PQ disturbance is really
relevant to be compensated and/or if their compensation is
economically feasible, rather than other solutions. Besides, since
the decision process is completely automatic, the inferences
are quite fast to be achieved and they are not dependent on the
availability of a high-qualied specialist.
Moreover, the DM output variables were designed in order to
associate different compensation architectures, each one oriented
to minimize specic PQ disturbances. This leads to a cooperative
compensation approach that may consider costly-effective solutions rather than simply installing the most sophisticated (and
expensive) devices.

E.V. Liberado et al. / Expert Systems with Applications 42 (2015) 35623570

Finally, the analysis of PQ indices related to both currents and

voltages expands the understanding about their relations to possible solutions and allows better choices when designing compensation systems, as depicted in case 6.
Some limitations of the proposed expert system would be: (1)
the denition of the time interval to make the measurements is
case-dependent and, if the load dynamic is not constant or known,
the characterization of all PQ disturbances would require longtime measurement campaigns; (2) the lack of standardization of
PQ indices and their limits, and the fact that most of the power
meters do not calculate all of them, would require the using of different instruments to acquire the relevant data or the employment
of specic PQ meter in order to embed or correlate the expert system in/to a single hardware.
Although the good results from using the proposed expert systems, future improvements are expected based on the inclusion of
additional and/or new power quality indices, in order to improve
the PQ characterization at the PCC and, the inclusion of compensators design procedures in the expert system, in such a way that the
expert system may not only suggest the compensation solution but
offer basic specication for the compensators (e.g., nominal power,
number of harmonic components to be compensated, etc.). Furthermore, some complementary information about the rated price
of each compensator may be included in the FRBS. Finally, the
implementation of the expert system on an embedded system, as
well as the use of the proposed approach in on-line applications
for smart grids (on which the expert system could manage the distribution of the compensation demand among available compensators), is expected.
Acknowledgements
The authors wish to express their gratitude to FAPESP (So
Paulo Research Foundation, Process 2012/14014-0), CNPq (Process
487471/2012-1) and Capes for supporting this research.
References
Akagi, H., Watanabe, E. H., & Aredes, M. (2007). Instantaneous power theory and
applications to power conditioning (1st ed.). Wiley-IEEE Press.
Arghandeh, R., von Meier, A., & Broadwater, R. (2014). Phasor-based approach for
harmonic assessment from multiple distributed energy resources. In
Proceedings of IEEE PES general meeting conference & exposition (p. 15).
Banerjee, A., Mukherjee, V., & Ghoshal, S. P. (2014). Intelligent fuzzy-based reactive
power compensation of an isolated hybrid power system. International Journal
of Electrical Power and Energy Systems, 57, 164177.
Bellman, R. E., & Zadeh, L. A. (1970). Decision-making in a fuzzy environment.
Management Science, Application Series, 17, B141B164.
Bhattacharya, S. K., & Goswami, S. K. (2009). A new fuzzy based solution of the
capacitor placement problem in radial distribution system. Expert Systems with
Applications, 36, 42074212.
Bisanovic, S., Hajro, M., & Samardzic, M. (2014). One approach for reactive power
control of capacitor banks in distribution and industrial networks. International
Journal of Electrical Power and Energy Systems, 60, 6773.
Bollen, M. H. J. (2003). What is power quality? Electric Power Systems Research, 66,
514.
Caia, Y. P., Huangc, G. H., Lina, Q. G., Niea, X. H., & Tana, Q. (2009). An optimizationmodel-based interactive decision support system for regional energy
management systems planning under uncertainty. Expert Systems with
Applications, 36, 34703482.
Chang, Y. (2010). Integration of SQP and PSO for optimal planning of harmonic
lters. Expert Systems with Applications, 37, 25222530.
Chang, Y., Low, C., & Hung, S. (2009). Integrated feasible direction method and
genetic algorithm for optimal planning of harmonic lters with uncertainty
conditions. Expert Systems with Applications, 36, 39463955.
Cheng, C. (2011). Design of output lter for inverters using fuzzy logic. Expert
Systems with Applications, 38, 86398647.
Cheng, P. T., Chen, C., Lee, T. L., & Kuo, S. (2009). A cooperative imbalance
compensation method for distributed-generation interface converters. IEEE
Transactions on Industry Applications, 45, 805815.
Chompoo-Inwai, C., & Mungkornassawakul, J. (2013). A smart recording power
analyzer prototype using lab view and low-cost data acquisition (DAQ) in being
a smart renewable monitoring system. In Proceedings of IEEE green technologies
conference (pp. 4956).

3569

Corasaniti, V. F., Barbieri, M. B., Arnera, P. L., & Valla, M. I. (2009). Hybrid active lter
for reactive and harmonics compensation in a distribution network. IEEE
Transactions on Industrial Electronics, 56, 670677.
Currence, E. J., Plizga, J. E., & Nelson, H. N. (1995). Harmonic resonance at a mediumsized industrial plant. IEEE Transactions on Industry Applications, 31, 682690.
Faisal, M., Mohamed, A., Shareef, H., & Hussain, A. (2011). Power quality diagnosis
using time frequency analysis and rule based techniques. Expert Systems with
Applications, 38, 1259212598.
Galvo, T. M., Belchior, F. N., Silveira, P. M., & Ribeiro, P. F. (2014). Comparative
analysis of power quality instruments measuring voltage and power. In
Proceedings of IEEE 16th international conference on harmonics and quality of
power (ICHQP) (pp. 762767).
Golovanov, N., Lazaroiu, G. C., & Porumb, R. (2013). Wind generation assessment
proposal by experimental harmonic and distortion factor analysis. In
Proceedings of 48th international universities power engineering conference
(UPEC) (pp. 14).
Gunal, S., Gerek, O. N., Ece, D. G., & Edizkan, R. (2009). The search for optimal feature
set in power quality event classication. Expert Systems with Applications, 36,
1026610273.
Gyugy, L., Otto, R. A., & Putman, T. H. (1978). Principles and applications of static,
thyristor-controlled shunt compensators. IEEE Transactions on Power Apparatus
and Systems, 97, 19351945.
Hashemi, S., & Aghamohammad, M. R. (2013). Wavelet based feature extraction of
voltage prole for online voltage stability assessment using RBF neural network.
International Journal of Electrical Power and Energy Systems, 49, 8694.
Hassan, L. H., Moghavvemi, M., Almurib, H. A. F., & Steinmayer, O. (2013). Current
state of neural networks applications in power system monitoring and control.
International Journal of Electrical Power and Energy Systems, 51, 134144.
IEEE Institute of Electrical and Electronics Engineers. (1993). IEEE recommended
practices and requirements for harmonic control in electrical power systems. In
IEEE Standard (pp. 5191992).
IEEE Institute of Electrical and Electronics Engineers. (2010). IEEE standard
denitions for the measurement of electrical power quantities under sinusoidal,
nonsinusoidal, balanced or unbalanced conditions. IEEE Std 1459-2010.
Jegathesan, V., & Jerome, J. (2011). Elimination of lower order harmonics in voltage
source inverter feeding an induction motor drive using evolutionary algorithms.
Expert Systems with Applications, 38, 692699.
Jintakosonwit, P., Srianthumrong, S., & Jintagosonwit, P. (2007). Implementation and
performance of an anti-resonance hybrid delta-connected capacitor bank for
power factor correction. IEEE Transactions on Power Electronics, 22, 25432551.
Kriukov, A., Grigoras, G., Scarlatache, F., Ivanov, O., & Vicol, B. (2014). Use of fuzzy
techniques in reliability assessment of electric distribution systems. In
Proceedings of IEEE 16th international conference on harmonics and quality of
power (ICHQP) (pp. 2933).
Liberado, E. V., Souza, W. A., Pomilio, J. A., Paredes, H. K. M., & Marafo, F. P. (2013).
Design of static var compensators using a general reactive energy denition.
Przeglad Elektrotechniczny (Electrical Review), 89, 233238.
Low, C., Chang, Y., & Hung, S. (2009). An application of sequential neural-network
approximation for sitting and sizing passive harmonic lters. Expert Systems
with Applications, 36, 29102920.
Marafo, F. P., Souza, W. A., Liberado, E. V., Silva, L. C. P., & Paredes, H. K. M. (2013).
Load analyser using conservative power theory. Przeglad Elektrotechniczny
(Electrical Review), 12, 16.
Morsi, W. G., & El-Hawary, M. E. (2008a). A new fuzzy-based representative quality
power factor for nonsinusoidal situations. IEEE Transactions on Power Delivery,
23, 930936.
Morsi, W. G., & El-Hawary, M. E. (2008b). A new fuzzy-based representative quality
power factor for unbalanced three-phase systems with nonsinusoidal
situations. IEEE Transactions on Power Delivery, 23, 24262438.
Obulesu, Y. P., Reddy, M. V., & Kusumalatha, Y. (2014). A %THD analysis of industrial
power distribution systems with active power lter-case studies. International
Journal of Electrical Power and Energy Systems, 60, 107120.
Paredes, H. K. M., Costabeber, A., & Tenti, P. (2011a). Application of conservative
power theory to cooperative control of distributed compensators in smart grids.
Przeglad Elektrotechniczny (Electrical Review), 87, 17.
Paredes, H. K. M., Silva, L. C. P., Brando, D., & Marafo, F. P. (2011b). Possible shunt
compensation strategies based on conservative power theory. Przeglad
Elektrotechniczny (Electrical Review), 87, 3439.
Phipps, J. K., Nelson, J. P., & Sen, P. K. (1994). Power quality and harmonics on
distribution system. IEEE Transactions on Industry Applications, 30, 476484.
Piltan, M., Mehmanchi, E., & Ghaderi, S. F. (2012). Proposing a decision-making
model using analytical hierarchy process and fuzzy expert system for
prioritizing industries in installation of combined heat and power systems.
Expert Systems with Applications, 39, 11241133.
Rahmani, S., Hamadi, A., Al-Haddad, K., & Dessaint, L. A. (2014). A combination of
shunt hybrid power lter and thyristor-controlled reactor for power quality.
IEEE Transactions on Industrial Electronics, 61, 21522164.
Ray, R. N., Chatterjee, D., & Goswami, S. K. (2010). A PSO based optimal switching
technique for voltage harmonic reduction of multilevel inverter. Expert Systems
with Applications, 37, 77967801.
Sahin, S., Tolun, M. R., & Hassanpour, R. (2012). Hybrid expert systems: A survey of
current approaches and applications. Expert Systems with Applications, 39,
46094617.
Salem, M., Mohamed, A., & Samad, S. A. (2010). Rule based system for power quality
disturbance classication incorporating S-transform features. Expert Systems
with Applications, 37, 32293235.

3570

E.V. Liberado et al. / Expert Systems with Applications 42 (2015) 35623570

Singh, B., Al-Haddad, K., & Chandra, A. (1999). A review of active lters for power
quality improvement. IEEE Transactions on Industrial Electronics, 46, 960971.
Su, H. J., Chang, G. W., Hsu, L. Y., Lu, H. J., Lee, Y. D., Chang, Y. R., & Lin, J. H. (2013).
Development of power quality analysis platform for INER Microgrid. In
Proceedings of IEEE power and energy society general meeting (PES) (pp. 15).
Suslov, K. V., Solonina, N. N., & Smirnov, A. S. (2014). Distributed power quality
monitoring. In Proceedings of IEEE 16th international conference on harmonics and
quality of power (ICHQP) (pp. 517520).
Tutkun, N. (2010). Improved power quality in a single-phase PWM inverter voltage
with bipolar notches through the hybrid genetic algorithms. Expert Systems with
Applications, 37, 56145620.
Valle, F., Klonari, V., Lisiecki, T., Durieux, O., Moiny, F., & Lobry, J. (2013).
Development of a probabilistic tool using Monte Carlo simulation and smart

meters measurements for the long term analysis of low voltage distribution
grids with photovoltaic generation. International Journal of Electrical Power and
Energy Systems, 53, 468477.
Wang, M., & Tseng, Y. (2011). A novel analytic method of power quality using
extension genetic algorithm and wavelet transform. Expert Systems with
Applications, 38, 1249112496.
Yang, Z., Cao, J., Xu, Y., Zhang, H., Yu, P., & Yao., S. (2013). Data Cleaning for Power
Quality Monitoring. In Proceedings of fourth international conference on
networking and distributed computing (ICNDC) (pp. 111115).
Zhang, J., Yan, W., Yang, Q., Bao, Z., & Sun, R. (2013). Power quality measurement
and analysis of offshore wind farm based on PSCAD/EMTDC models. In
Proceedings of sixth international conference on advanced computational
intelligence (ICACI) (pp. 371375).