Electric Power Systems Research 81 (2011) 1079–1087

Contents lists available at ScienceDirect

Electric Power Systems Research

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

Distribution network phase load balancing as a combinatorial optimization

problem using fuzzy logic and Newton–Raphson
Willy Mukwanga Siti ∗ , Adisa Jimoh ∗∗ , Dan Nicolae 1
Department of Electrical Engineering, Tshwane University of Technology, Private Bag X680, Pretoria 0001, South Africa

a r t i c l e i n f o a b s t r a c t

Article history: The electrical network system is to ensure that an adequate supply is available to meet the estimated
Received 19 July 2009 load of the consumers in both the near and more distant future. In order to avoid excessive voltage
Received in revised form 4 December 2010 drop and minimize loss, it may be economical to install apparatus to balance or partially balance the
Accepted 15 December 2010
loads. It is believed that the technology to achieve an automatic load balancing lends itself readily for
Available online 21 January 2011
the implementation of different types of algorithms for automatically rearranging the connection of
consumers on the low voltage end of a feeder for optimal performance. In this paper the combination of
Keywords:
the fuzzy logic, which addresses the linear solution, with Newtown Raphson as optimization method for
Load balancing
Fuzzy logic
the nonlinear behavior of the load is developed. A study case is presented for a radial feeder of 150 loads.
Newton–Raphson
© 2010 Elsevier B.V. All rights reserved.

1. Introduction sequently results into service quality and operation efficiency being
reduced [2]. Phase unbalance is also manifested in increased com-
Energy efficiency in the electricity supply system has always plex power unbalance, increased power loss, enhanced voltage
being a concern. With the increase in electric power consump- drop, and increased neutral current.
tion and the advancement in the distribution system technology In most of the cases, the phase voltage and current unbal-
is the challenge to further decrease the system technical loss, ances can be greatly improved by suitably arranging the connection
and improve performance. To meet this challenge, automation of phases between the distribution transformers and a primary feeder.
the power distribution system needs to be widely adopted. All It is also possible to advance the phase current unbalances in every
switches and circuit-breakers involved in the controlled networks feeder segment by means of changing the connection phases [3].
are equipped with facilities for remote operation. The control inter- The phase voltage unbalances along a primary feeder can also
face equipments must withstand extreme climatic conditions. Also, be improved in common cases by system reconfiguration, which
control equipments at each location must have a dependable power involves the rearrangement of loads or transfer of load from heavily
source. To cope with the complexity of the electricity distribu- loaded area to the less loaded.
tion, the latest computer, communication, and power electronics In the modern power distribution systems, the sectionalizing
equipment in distribution technologies are needed to be employed. switches and the tie switches for primary feeder reconfigura-
The distribution automation can be defined as an integrated sys- tion are extensively used [2,3]. The authors in [4,5] presented
tem concept [1]. It includes control, monitoring and, sometimes, the way to control the tie switches using heuristic combinatorial
decision to alter any kind of loads. The automatic distribution sys- optimization-based method. The only disadvantage with the tie-
tem provides directions for automatic reclosing of the switches switch control is that, in most of the cases, it makes the current
and remote monitoring of the loads contributing towards phase and the voltage unbalances worse. In the attempt to overcome
balancing. this shortcoming other researchers [6,7] have tried to evolve new
The phase voltage and current unbalances are major factors approaches to the solution of the problem. Reference [8] presented
leading to extra losses, communication interference, equipment the use of the artificial neural networks to find the optimum switch-
overloading and malfunctioning of the protective relay which con- ing option of the loads among the different phases. On the basis of
these results, other networks identify the radial topology satisfying
the optimal condition.
The focus of this paper, however, is not about reconfiguration
∗ Corresponding author. Tel.: +27 824961577.
∗∗ Corresponding author. Tel.: +27 123824964; fax: +27 123824964. through feeder bifurcation and the tie and sectionalizing switches
E-mail addresses: [email protected] (W.M. Siti), [email protected] (A. Jimoh). controls, but phase balancing of the secondary feeder by rearrang-
1
Tel.: +27 826797483. ing the connection to consumer loads. In this the phase balancing

0378-7796/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2010.12.006
1080 W.M. Siti et al. / Electric Power Systems Research 81 (2011) 1079–1087

Fp1 is three-phase, four-wire system that accommodates three-phase

and single-phase consumer loads. The central computing unit
(CCU) takes information from the loads connected to the secondary
feeders under surveillance about consumption and transmits con-
Fp2
trol signal to the load selector switches (LSS). For the problem in
this paper, it is assumed that each feeder contains n × m domestic
loads.
Fpl As shown in Fig. 1(b) supposing each of the n secondary feeders
has connected to it m loads, and each load can be connected through
Fs1 Fsk Fsn
the switch selector to only one of the three phases. The status of
each of the loads connected to an unbalanced feeder is sent through
CCU a communication interface to a load re-arrangement/balancing
algorithm. In turn the algorithm returns, after processing, a re-
arrangement status, through the same communication interface,
that should be implemented to yield a balanced feeder arrange-
ment.
To elucidate further, Fig. 2 shows an individual switch selector
LSS Lkj for a random load Lkj (load j on feeder k) and its communication
with the automation controller. The central unit receives informa-
tion of currents/power {[I1 I2 I3 ]k [P1 P2 P3 ]k } from feeders but also
from individual loads Lkj {Ikj Pkj }. Indices 1, 2 and 3 represent phase
Ph1 Ph2 Ph3 1, phase 2 and phase 3, respectively. After computing the optimum
a switching matrix configuration, the information is transmitted to
the load switch selectors {[sw]kj } to move or not the respective load
from one phase to another in order to achieve balancing. The com-
F {[I] [P] F {[sw] }
puting of the new switching matrix configuration will stop when
the unbalancing between phases is lower than a preset threshold.
Communication Interface

Communication Interface

The transfer of the load from one phase to another is done via
F {[I] [P] } F {[sw] } the three bilateral thyristors/triacs such that at any moment in time
Balancing
Algorithm only one switch is ON to prevent phase short circuit. This is the main
set of constrains of the system:

swkj1 + swkj2 + swkj3 − 1 = 0 (1)

F {[I] [P] } F {[sw] }
Automation System
where swkj1 , swkj2 and swkj3 can be only 1 or 0 (conducting or not).
The status of the gate matrix is provided from the main con-
troller. Actual transition from one phase to another is done at zero
b crossing to avoid voltage spikes. This in return will introduce a dip
Fig. 1. General representation of the proposed automated distributed system.
of maximum 17 msec (see Fig. 3) which is acceptable by the power
quality standards.
For example, the decision to reconfigure is taken at moment A in
problem is mathematically formulated and then solved. The results Fig. 3. The zero crossing switches OFF the load current (iload ) from
are therefore used to initiate certain actions to eliminate or min- phase 1 at moment B and switches ON again on phase 2 at moment
imize the problem. With the fuzzy logic it is wondered if there C.
cannot be simpler, more straightforward, better, and faster solution
method. It is believed that the technology to achieve an auto- 2.2. Problem definition
matic load balancing lends itself readily for the implementation
of different types of algorithms for automatically rearranging the The network reconfiguration problems for loss reduction and
connection of consumers on the low voltage end of a feeder for opti- load balancing involve load transfer between the feeders or substa-
mal performance. In this paper the combination of the fuzzy logic tions by changing the position of switches. In achieving this, some
with Newton–Raphson as optimization method is implemented to predefined objectives has to be satisfied under different constraints.
balance the load in the secondary part of the transformer. Consumer loads are mostly single phase, and these single phases
are arranged or grouped, under an ideal situation, to have a bal-
2. Problem description anced three phase systems, however, unbalance still occurs due
to the unequal load distribution among the phases of the feeder.
2.1. Proposed technique Hence the need to minimize the unbalance by transferring loads
from the heavily loaded to the less loaded phases so as to reduce
The low voltage distribution system is formed from feeders that power loss, voltage drop etc.
usually are three-phase, four-wire system with a radial or open In formulating this problem, the redistribution of load among
loop structure. To improve the system phase voltage and current the phases must be such that certain predefined objectives are
unbalances the connection between the specific feeder and the dis- satisfied. In this work the objective functions that are considered
tribution transformer should be suitably arranged. The domestic are:
loads are connected, as in most cases, in a single-phase. Fig. 1(a)
shows a typical radial distribution system with l primary feeders 2.2.1. Total complex power unbalanced
Fp1 , Fp2 , . . .Fpl . From each primary feeder are n secondary feed- In a three phase load system, the complex powers for a typical
j
ers to which consumer loads are connected. Each secondary feeder feeder i are expressed as s̄i (j = 1, 2 and 3) for the loading of each
 W.M. Siti et al. / Electric Power Systems Research 81 (2011) 1079–1087 1081

F1 {[I]1 [P]1}
CCU Load Switching Selector
Ph1

Ph2

Communication Interface

Communication Interface
Fk {[I]k [P]k} Ph3
swkj1 swkj2 swkj3
Balancing
Algorithm

Lkj [sw]kj

Interface
Control
Ikj (Pkj)
Fn {[I]n [P]n} O

Lkj

Fig. 2. General description.

A B C where Pph1 , Pph2 and Pph3 are the loads (power) drawn from the
400 phase one, two and three, respectively. AU is the average power
Ph. 1 Ph. 2 Ph. 3 unbalance to be minimized.
300
Relative amplitude (V, A)

200
2.2.2. Phasor current unbalance relationship
100 In a three phase unbalanced system the objective is to minimize
the difference of the amplitude of the phase currents (Îphik ):
0
 
-100
 IPh1k − IPh2k 
iload Minimize J =  IPh1k − IPh3k  (6)
 
-200 I −I 
Ph2k Ph3k
-300
where Iph1k , Iph2k and Iph3k represent the currents (phasors) for each
-400 of the phases (1, 2 and 3) at an arbitrary point of connection k.
0.15 0.16 0.17 0.18 0.19 0.2 0.21 0.22

Fig. 3. Load transition from one phase to another. 2.2.3. Neutral current of an unbalanced circuit
The current at the neutral point of a balanced three phase system
should be 0. Therefore, the objective in balancing an unbalanced
phase [13]. The unbalance of these three complex powers siul can system is to minimize the neutral current:
be evaluated as follows: 2 2 2 2
 Minimize Jn = [IN2 − ((Iph1k − Iph2k ) + (Iph1k − Iph3k ) + (Iph2k − Iph3k ) )] (7)
 1 c j id
  
ul
si =  |s̄ − s̄ |2 i i
(2) 2.2.4. Zero and negative sequence currents of an unbalanced
3
j=a circuit
For a balanced three-phase circuit the zero and negative
where s̄iid represents the ideal per phase loading and: sequence components are zero. Therefore, the objective is to min-
imize the negative and zero sequence currents:
c
1
j 0 −
s̄iid = s̄i (3) Minimize Jon = Iphs + Iphs (8)
3
j=a
where  is the Lagrange multiplier.
Thus an evaluation of s̄iul = 0 indicate that the complex power For the real implementation of a control system, the following
of ith feeder is balanced. The total complex power unbalance can elements are necessary:
be evaluated from:
• A measurement system for real loads.
n
 • A transmission data system for the load data connecting each
sTul = s̄iul (4)
point.
i=1 • A transmission system for sending the input signals to the switch
where n is the number of feeders within the object feeder. The term breaker.
sTul is used to evaluate the complex power unbalance of a feeder and • The control cannot start if the above described components and
the value of this describes the load balance situation at every feeder. system are not properly installed and in correct condition.
Eq. (2) can be simplified, for the real power only, to obtain the
following as the main objective function applied in this study: 2.3. Proposed algorithm

AU (|Pph1 − Pph2 | + |Pph2 − Pph3 | + |Pph3 − Pph1 |) The solution objective for a feeder is to obtain a set of rear-
= (5)
ph 3 rangement of the connected loads at each node (or consumer point)
1082 W.M. Siti et al. / Electric Power Systems Research 81 (2011) 1079–1087

Phase
Load

Is Avg.
Unbalance No Change
/phase > Final Load
10 kW ?

Fig. 5. Fuzzy membership functions for the input variable.

Fuzzy-based cific phase is less loaded and should receive that amount of
Load Balancing load.
This load change configuration is the input to the expert system
which tries to optimally shift the specific number of load points.
However, sometimes the expert system may not be able to execute
the exact amount of load change as directed by the fuzzy step. This
Newton Raphson is because the actual load points for any phase might not result
Combinatorial in an optimum combination which sums up to the exact change
Optimization value indicated by the fuzzy step. So, the best possible change from
the expert system is implemented, and iteratively check the sys-
tem unbalance until the average unbalance (AU) below 10 kW is
achieved.

3. Fuzzy logic-based load balancing

Fig. 4. Architecture for the proposed load balancing system.

In this section, the fuzzy logic-based load balancing technique

such that the objective function is minimized. This is a non-linear is described in details. Further a set of 150 loads is considered as a
problem that will involve a number of trial and errors. It is hereby case study, see Table 1. As generally accepted the power factor of a
proposed to solve this problem as a combinatorial optimization, house hold is very close to unity and for this reason the load power
where iteratively, as changes are made, a method is used to sense is considered to be active power. For this case study the average
the relative loading of the phases, and another method is used to per phase capacity of the system is 300 kW.
edge towards the minimized objective. As can be clearly seen phase one is overloaded which can pro-
The objective functions are formulated as fuzzy sets since they duce overheating of the transformer. The other phases are less
are imprecise. Fuzzy logic (FL) has two different meanings. In a nar- loaded and then the need for rearrangement to attain balancing. The
row sense, fuzzy logic is a logical system, which is an extension of average unbalance per phase according to Eq. (4) in this situation
multi-valued logic [9–12]. However, in a wider sense fuzzy logic is: AU = 109 kW.
(FL) is almost synonymous with the theory of fuzzy sets, a the- For designing the fuzzy controller, it is further assumed that the
ory which relates to classes of objects with unsharp boundaries in maximum overload capacity of any phase to be 300 kW. Beyond
which membership is a matter of degree. In this perspective, fuzzy 300 kW the fuzzy controller should not be used for load balancing.
logic in its narrow sense is a branch of FL. Even in its more narrow
definition, fuzzy logic differs both in concept and substance from 3.1. Fuzzy controller: input and output
traditional multivalued logical systems.
For the above-mentioned system, in this paper, it is pro- To design the fuzzy controller, there first are specified the
posed a fuzzy logic-based load balancing technique along with the input and the output variables. For the load balancing purpose, as
Newton–Raphson method that is solving the objective function as described in Section 2.3, it is chosen the input as ‘Load’, i.e., the
a combinatorial problem. The architecture of the proposed algo- total phase load (kW) for each of the three phases, and the output
rithm is shown in Fig. 4. In Fig. 4, the input is the total phase load as ‘Change’, i.e., the change of load (kW, positive or negative) to be
(for each of the three phases). The average unbalance (AU) per made for each phase. For the input variable, Fig. 5 and Table 2 show
phase, calculated according to Eq. (4), is checked against a spec- the fuzzy nomenclature and the respective triangular fuzzy mem-
ified threshold of say 10 kW. If the average unbalance per phase is bership functions [9]. And for the output variable, Table 3 shows the
below 10 kW, it can be assumed that the system is more or less bal- fuzzy nomenclature and Fig. 6 the corresponding triangular fuzzy
anced and discard any further load balancing. Otherwise, it goes for membership functions [9,10].
the fuzzy logic-based load balancing. This 10 kW threshold is arbi-
trarily/conveniently chosen in order to reduce the rate of system 3.2. Fuzzy rules and surface
reconfiguration.
The output from the fuzzy-based load balancing step is the Next, it is determined the IF-THEN fuzzy rule set [9,10] govern-
load change values for each phase. A negative value indicates ing the input and output variable as described in Table 4.
that the specific phase has surplus load and should release that Corresponding to the fuzzy input, output variables and the asso-
amount of load, while a positive value indicates that the spe- ciated rule set, the fuzzy surface [9] is shown in Fig. 7, depicting
 W.M. Siti et al. / Electric Power Systems Research 81 (2011) 1079–1087 1083

Table 1
Study case.

Phase 1 Phase 2 Phase 3

Load kW Load kW Load kW

L1 7.4 L51 2.3 L101 4.9

L2 2.7 L52 0.1 L101 3.2
L3 3.2 L53 0.1 L103 3.1
L4 6.1 L54 2.3 L104 0.1
L5 6.0 L55 2.7 L105 0.1
L6 6.1 L56 2.6 L106 1.8
L7 3.7 L57 0.1 L107 2.2
L8 8.1 L58 2.0 L108 0.1
L9 7.2 L59 0.1 L109 0.1
L10 2.6 L60 2.1 L110 0.1
L11 1.7 L61 3.0 L111 0.1
L12 5.1 L62 1.7 L112 0.1
L13 1.0 L63 3.3 L113 2.5
L14 4.6 L64 1.7 L114 1.8
L15 4.7 L65 1.5 L115 3.0
L16 5.6 L66 3.1 L116 6.5
L17 5.2 L67 0.1 L117 0.1 Fig. 7. Nonlinear relationship between the input and the output variable.
L18 2.9 L68 0.1 L118 0.1
L19 5.0 L69 3.4 L119 1.8
L20 3.0 L70 6.0 L120 3.1
L21 6.9 L71 3.0 L121 0.1
the nonlinear relationship between the input and the output
L22 5.1 L72 4.2 L122 2.0
L23 7.9 L73 4.8 L123 1.8 variable.
L24 4.8 L74 2.6 L124 3.0
L25 4.3 L75 4.6 L125 1.6 4. Newton–Raphson optimization
L26 4.7 L76 2.5 L126 0.5
L27 5.5 L77 4.5 L127 2.2
L28 4.6 L78 0.1 L128 1.8 Till this point, the fuzzy logic part of the algorithm treated the
L29 5.2 L79 0.1 L129 1.4 system as a linear one. But, as can be seen in Fig. 8, the load is
L30 4.9 L80 2.6 L130 0.8 non-linear. Due to this non-linearity, further processing technique
L31 4.8 L81 4.6 L131 2.5
requires Newton–Raphson optimization.
L32 4.6 L82 2.5 L132 0.1
L33 4.8 L83 4.7 L133 0.1
L34 4.9 L84 2.4 L134 0.1
Table 2
L35 4.2 L85 0.1 L135 0.1
Fuzzy nomenclature for the input variable.
L36 7.2 L86 0.1 L136 0.1
L37 4.6 L87 2.4 L137 0.1 SL. no. Input (load) description Fuzzy nomenclature kW range
L38 5.2 L88 2.7 L138 0.1
L39 4.7 L89 4.5 L139 6.5 1 Very less loaded VLL 0–50
L40 5.0 L90 4.3 L140 1.3 2 Less loaded LL 35–85
L41 5.1 L91 0.1 L141 0.5 3 Medium less loaded MLL 65–115
L42 5.2 L92 0.1 L142 1.3 4 Perfectly loaded PL 100–150
L43 5.6 L93 2.7 L143 1.6 5 Slightly overloaded SOL 125–175
L44 4.4 L94 4.5 L144 0.1 6 Medium overloaded MOL 165–215
L45 4.6 L95 3.4 L145 3.5 7 Overloaded OL 200–250
L46 5.2 L96 2.2 L146 0.1 8 Heavily overloaded HOL 235–300
L47 5.6 L97 7.2 L147 2.9
L48 5.9 L98 0.1 L148 0.1
L49 3.2 L99 2.4 L149 4.3 Table 3
L50 5.2 L100 0.1 L150 0.1 Fuzzy nomenclature for the output variable.
P1 245 P2 120 P3 82
SL. no. Output (change) description Fuzzy nomenclature kW range

1 High subtraction HS −150 to −85

2 Subtraction S −100 to −50
3 Medium subtraction MS −65 to −15
4 Slight subtraction SS −50 to 25
5 Perfect addition PA 0–50
6 Medium addition MA 35–85
7 Large addition LA 65–115
8 Very large addition VLA 100–150

Table 4
Fuzzy rules for the input and output variable.

Rule no. Rule description

1 If load VLL then change is VLA

2 If load is LL then change is LA
3 If load is MLL then change is MA
4 If load is PL then change is PA
5 If load is SOL then change is SS
6 If load is MOL then change is MS
Fig. 6. Fuzzy membership functions for the output variable. 7 If load is OL then change is S
8 If load is HOL then change is HS
1084 W.M. Siti et al. / Electric Power Systems Research 81 (2011) 1079–1087

+
Iph1k + aIph2k + a2 Iph3k
Iphs = (15)
3

−
Iph1k + a2 Iph2k + aIph3k
Iphs = (16)
3
Applying the Lagrange multiplier for the constraint to balance
the three phase system results in:

0
(Iph1k + Iph2k + Iph3k )
Iphs = (17)
3

+
(Iph1k + aIph2k + a2 Iph3k )
Iphs = (18)
3

−
(Iph1k + a2 Iph2k + aIph3k )
Iphs = (19)
Fig. 8. Determination of the output load change for phase 1 of the input load. 3
To have a balanced distribution feeder the equations presented
In this optimization process, the neutral current gives immedi- as constraints Eqs. (17) and (18) will be equal to zero. The com-
ate information about unbalancing of the phase currents/powers. bination of Eqs. (6), (7), (17) and (18) can be integrated as a
Hence, the currents will be used into the Newton–Raphson opti- multi-objective optimization problem as follows:
mization process. 0 −
Minimize Joj = J + Jn + Iphs + Iphs (20)
Given a distribution system as shown in Fig. 1, a net-work with
3 phases with a known structure, the problem consists of finding a To solve the minimization problem, the gradient of the least
condition of balancing. In any connecting point, the mathematical square objective function Joj as defined in Eq. (20), can be expressed
model can be expressed as: in terms of x and set equal to zero.
3
 
 ∂Joj ∂Joj ∂Joj
Iph1k = swkji Ik + Iph1(k+1) (9) Jx = , , =0 (21)
∂swk1i ∂swk2i ∂swk3i
i=1

3
where x = [swki , swki , swki ] is the vector of the switching matrix.
 The Eq. (23) is a system of non linear equations. To solve the
Iph2k = swkji Ik + Iph2(k+1) (10)
system of non linear equations, the equation system is linearized
i=1
around some working points xk by using a truncated Taylor series
3
 expansion:
Iph3k = swkji Ik + Iph3(k+1) (11)
Jxx (xk )xk + Jx (xk ) = 0 (22)
i=1
where Jxx (xk ) is the 3 × 3 Hessian matrix, containing the second
where Iph1k , Iph2k and Iph3k represent the currents (phasors) per
order derivatives of the objective function Joj evaluated at point xk ,
phase (1, 2 and 3) after the k – an arbitrary point of connection; swkj1
and Jx (xk ) is the gradient of J evaluated at point xk . The correction
. . . swkj3 are different switches (the value of ‘1’ means the switch is
vector xk can then be calculated by solving the following system
ON and ‘0’ means it is OFF). Iph1(k+1) , Iph2(k+1) and Iph3(k+1) represent
of linear equations:
the currents (phasors) per phase (1, 2 and 3) after the previous point
of connection (conventionally, the first point is at the front end of Jxx (xk )xk = −Jx (xk ) (23)
the feeder). Balancing the phase current of any feeder, the neutral
Initially, an arbitrary value is given to the parameter vector xk
current should be minimized and consequently the power losses
and then an iterative procedure is used to obtain a better value of
associated diminished. Therefore, the second objective is to mini-
the parameter vector.
mize the difference of the amplitude of the phase currents (Îphik ) as
presented in Eq. (6).The third objective function is to minimize the xk+1 = xk + xk (24)
neutral current, which can be expressed as in Eq. (7).
The fourth objective functions will be to minimize the negative 5. Simulation result
and zero sequence currents. According to Fortescue [16] the three
unbalanced phasors of a three-phase system can be resolved into In this section, the application results using the fuzzy logic-
three balanced systems phasors known as positive, negative and based load balancing technique is shown. Matlab® fuzzy toolbox
zero sequence components. The relation between the symmetrical [9] was used for the simulation. Mamdani [11] fuzzy inference
components and the phase current are: technique has been utilized. Let us take for an example input load
Iphs = A−1 Iphk (12) configuration of [24512082] kW for the three phases. The above
described fuzzy controller is then used to do the balancing. The
and A represent the main matrices. Then Eq. (12) becomes: graphical determination of the output load change for the three
⎡ 0
IPhs
⎤   phases corresponding to this input load and involving the eight
1 1 1 Iph1k
+
⎣ Iphs ⎦= 1 1 a a2 Iph2k (13)
fuzzy rules are shown in Figs. 8 and 9. The defuzzification operation
3 is based on the Mamdani (centroid) technique [11].
−
Iphs 1 a2 a Iph3k
Where the first part of Fig. 8 shows the really unbalance in phase
Writing as three separate equations, where Iphs 0 , I + , I − are, 1, and the second part shows the subtraction from heavily loaded
phs phs
to less loaded feeder. This graphical representation is extended to
respectively, the zero sequence current, the positive sequence cur-
Fig. 9 as well, for the representation in phase 3.
rent and the negative sequence currents.
Fig. 8 shows a graphic representation of the triangular fuzzy
0
Iph1k + Iph2k + Iph3k representation membership and how the actual load is accommo-
Iphs = (14)
3 dated. The first part (left) of Fig. 8 shows the input membership and
 W.M. Siti et al. / Electric Power Systems Research 81 (2011) 1079–1087 1085

Table 5
The comparison of the system before and after the balance.

Original After balancing

J (%) 0.8688 0.00084

Jn (%) 0.5342 0.00123
0
IPhs (%) 0.6833 0.6777
−
IPhs (%) 0.65919 0.2317

AE = −5 kW, Perror = [−5 −5 −5]T kW and:

  
−104 −5 −99
P = 25 − −5 = 30 (kW) (31)
65 −4 69
  
245 −99 146
Pfinal = Pin + P = 120 + 30 = 150 (kW) (32)
82 69 151
Fig. 9. Determination of the output load change for phase 3 of the input load.
Applying Eq. (1) on Pin and Pfinal , the initial absolute unbalance
(IAUB)/phase and final absolute unbalance (FAUB)/phase will be,
how the load of phase one fits in membership classification (see respectively:
Table 2). It shows that it is not only over loaded (membership 7),
but also heavily overloaded (membership 8), where the actual load IAUB
= 108.67 kW (33)
being close to 300 kW, which is the maximum overload capacity. Phase
The second part of Fig. 9 shows the output membership and how FAUB
= 3.33 kW (34)
the changes in the phase one load should be done. It results that Phase
the changes in the phase one require changes of −104 kW which
The reduction of unbalance indicates improvement of the phase
are interpreted as large and very changes (output membership 7
balancing. Table 5 shows the result of Newton–Raphson optimiza-
and 8). The bottom graph on the right hand side of the Fig. 9 shows
tion process upon the multi-function Eq. (20), before and after the
result in output change.
phase balancing.
The same graphic explanation is extended for phase two, only
These percentage results have a correspondence in neutral cur-
now the result is 24.1 kW change/adding which is considered as a
rent of 0.5 A and power unbalance of 0.25 kW. This proves the
slight subtraction/adding (output membership 4).So, after round-
efficiency of Newton–Raphson optimization technique.
ing the output load change, for the input load
The input loads for this study were acquired from a load data
T survey in a South African city [4]. The input superset consists of
Pin = [ 245 120 82 ] (kW) (25)
many load data for the three phases for any specific time-period of
the output load change configuration is: a day over a month. We selected the input loads randomly from the
T
superset. The results presented in Table 4 are chosen to represent
Pfuzzy = [ −104 25 65 ] (kW) (26) different phase loading conditions. For each application, the final
fuzzy output change configuration is passed onto the expert sys-
However, with this load change configuration, it is going to have
tem for implementing the load change operation, described in the
an error. Because the positive and the negative totals are not equal,
following section. Table 6 shows the reconfiguration of the loads
i.e., Pfuzzy = −14 = / 0 kW. So, if this load change configuration is
(phase connections) after balancing.
implemented, this will result in reduction of −14 kW of total load.
As can be noticed, after applying only the fuzzy logic part of the
This is not possible, because, with the load balancing it can only
algorithm, there is still a small unbalance below the threshold of
interchange the load points among the three phases, keeping the
10 kW. When the Newton–Raphson further optimization has been
total load same, i.e., without increasing or decreasing the total load.
used, the balancing is quite perfect. However, a too tight balanc-
So, an error correction [6] has to be performed. The average error
ing increases the computing power and the rate of reconfiguration
(AE) is given as:
(Table 7).
 1   
AE = round Pfuzzy (27)
3 6. Practical feasibility and economic considerations
Then, this average error is used to construct the error matrix
Perror , by distributing the AE evenly among the three phases. The technologies for the implementation of the scheme pro-
 posed in Section 2.1 of this paper are already in existence. They
AE comprise, basically, the following, which nowadays are common
Perror =  AE (28) feature in telecontrol and smart grid systems:
Pfuzzy − 2 ∗ AE
• Metering sensors,
Finally, the load change configuration P, is obtained by sub- • Wireless data communication, such as Wi-Fi,
tracting the Perror from the uncorrected fuzzy output Pfuzzy . • Programmable single chip microcomputer or microcontroller,
P = Pfuzzy − Perror (29) • Actuators, and
• Static power electronics switches with soft switching.

P = 0 (30)
These technologies are well proven for applicability within the
Applying Eqs. (29) and (30) [6] in this study case, the following described scheme proposed here; even though they are contin-
parameters are: uously being further developed for ruggedness, robustness and
1086 W.M. Siti et al. / Electric Power Systems Research 81 (2011) 1079–1087

Table 6 • Overstressing cables and transformers

Study case. • Poor service quality and operation
Load Ph Load Ph Load Ph

L1 2 L51 2 L101 3 All these are undesirable as lower the overall energy and service
L2 1 L52 2 L101 3 efficiency, and have serious economic and financial consequences
L3 2 L53 1 L103 3 to the utility.
L4 1 L54 2 L104 3 Currently most utilizes minimize this problem of unbalance
L5 1 L55 2 L105 3
in the low voltage feeders by changing the connection phase of
L6 2 L56 2 L106 3
L7 1 L57 2 L107 3 some feeders manually after some field measurement and analysis.
L8 1 L58 2 L108 3 This manual correction normally will require the use of addi-
L9 2 L59 2 L109 3 tional manpower, service interruption, and ensuring additional
L10 1 L60 2 L110 3
safety measures. Although in some cases this process can improve
L11 1 L61 2 L111 3
L12 2 L62 1 L112 3 the phase current balance, this strategy is more time consuming,
L13 2 L63 2 L113 3 requires supply interruption, unsafe and only last for a while before
L14 1 L64 2 L114 3 the process is repeated again [1].
L15 3 L65 2 L115 3 The proposed scheme here eliminates manual correction of
L16 3 L66 2 L116 3
phase unbalance, thus eliminating its limitations, including addi-
L17 1 L67 2 L117 3
L18 1 L68 2 L118 3 tional cost overheads and technical shortcomings. Furthermore, the
L19 3 L69 2 L119 3 technical problems listed above and their economic and financial
L20 1 L70 2 L120 3 consequences are addressed as well. Since the proposed scheme
L21 3 L71 2 L121 3
does not entail any running costs, the capital cost of putting the
L22 1 L72 2 L122 3
L23 3 L73 2 L123 3
scheme in place are easily paid off within short period of time by
L24 3 L74 2 L124 3 the savings in the correction of the technical consequences and
L25 1 L75 2 L125 3 overhead costs involved with the manual solution.
L26 1 L76 1 L126 3
L27 1 L77 2 L127 3
L28 1 L78 1 L128 3
7. Conclusions
L29 3 L79 2 L129 1
L30 1 L80 2 L130 3 In this paper, a fuzzy logic and the Newton–Raphson based load
L31 1 L81 2 L131 3 balancing system has been presented. The proposed solution uses
L32 1 L82 2 L132 1
the fuzzy logic to address the linear behavior of the system while
L33 3 L83 2 L133 3
L34 3 L84 2 L134 3 Newton–Raphson tackles the optimization of the non-linear aspect
L35 1 L85 2 L135 3 of the load.
L36 1 L86 2 L136 3 The load balancing system is tested at the three-phase, four-wire
L37 1 L87 2 L137 3 unbalanced feeders and the simulation results obtained on the Mat-
L38 3 L88 2 L138 3
L39 1 L89 2 L139 3
lab platform show a substantial improvement of the unbalanced
L40 1 L90 2 L140 3 conditions.
L41 1 L91 2 L141 3 The proposed phase balancing system using the fuzzy logic with
L42 3 L92 2 L142 3 the combination of the Newtown–Raphson is effective for reducing
L43 1 L93 2 L143 3
the feeder unbalance and can be generically further extended to
L44 1 L94 2 L144 3
L45 1 L95 2 L145 3 other distribution systems and feeder load configurations.
L46 1 L96 2 L146 3
L47 3 L97 2 L147 3 References
L48 1 L98 2 L148 1
L49 3 L99 2 L149 3 [1] D.C. Walters, G.B. Schele, Generic algorithm solution of economic dispatch
L50 1 L100 2 L150 3 with valve point loading, IEEE Transactions on Power Systems 8 (3) (1991)
1325–1331.
[2] M.E. Baran, W. Kelly, State estimation for real time monitoring of distri-
bution systems, IEEE Transactions on Power Systems 9 (3) (1994) 1601–
reliability, which are considered the major challenges of the prac-
1609.
tical implementation of the idea. [3] S. Cinvalar, J.J. Grainger, H. Yin, S.S.H. Lee, Distribution feeder reconfiguration
Several literatures [2–5,14,15] have elucidated the technical and for loss reduction, IEEE Transactions on Power Delivery 3 (3) (1988) 1217–
economic consequences of phase unbalance in the low voltage side 1223.
[4] M.W. Siti, D.V. Nicolae, A.A. Jimoh, A. Ukil, Reconfiguration and load balanc-
of the distribution networks: ing in the LV and MV distribution networks for optimal performance, paper
TPWRD-00418-2006, IEEE Transactions on Power Delivery 22 (4) (2007) 2534–
• 2540.
Real power loss
[5] M. Siti, A. Jimoh, D. Nicolae, Automatic Load balancing in Distribution Feeder,
• Voltage drop IECON 05, Raleigh, NC, USA, 2005.
• Communication interference [6] M.A. Kashem, G.B. Jasmon, V. Ganapathy, A new approach of distribution system
• Equipment overloading and malfunctioning reconfiguration for loss minimization, Electrical Power and Energy Systems 22
(2000) 269–276.
• Malfunction of the protective relay [7] D. Das, Reconfiguration of distribution system using fuzzy multi-
objective approach, Electrical Power and Energy Systems 28 (2006) 331–
338.
Table 7 [8] A. Ukil, W. Siti, J. Jordaan, Feeder load balancing using neural network, Lecture
Summary of the results. Notes in Computer Science 3972 (2006) 1311–1316.
[9] B. Kosko, Fuzzy Engineering, Prentice Hall, 1999.
Original Fuzzy only Fuzzy + Newton–Raphson [10] MATLAB® Documentation – Fuzzy Logic Toolbox, Version 6.5.0.180913a
P1 (kW) 245 146 148.63 Release 13, The Mathworks Inc., Natick, MA.
[11] E.H. Mamdani, Applications of fuzzy logic to approximate reasoning using lin-
P1 (kW) 120 150 149.00
guistic synthesis, IEEE Transactions on Computers 26 (12) (1977) 1182–1191.
P1 (kW) 82 151 149.37
[12] M. Sugeno, Industrial Applications of Fuzzy Control, Elsevier Science Pub. Co.,
IN (A) 638 40 0.5
1985.
 W.M. Siti et al. / Electric Power Systems Research 81 (2011) 1079–1087 1087

[13] J.-S.R. Jang, C.-T. Sun, Neuro-Fuzzy and Soft Computing: a Computational [15] M. Siti and A. Jimoh, Distribution Circuit Loss Minimization through Feeder
Approach to Learning and Machine Intelligence, Prentice Hall, 1997. Reconfiguration, SAUPEC 2004 Conference 2004, Stellenbosch, South Africa.
[14] T. Chen, J. Cherng, Optimal phase arrangement of distribution connected to [16] C.L. Fortescue, Method of Symmetrical Co-Ordinates Applied to solution of
a primary feeder for system unbalance improvement and loss reduction using polyphase Networks, 34th Annual convention of the AIEE, July 1918 AIEE Trans-
genetic algorithms, IEEE Transactions on Power Systems 15–3 (2000) 994–999. actions Vol. 37 part II, 1918, pp. 1027–1140.