1

An Analytical Approach for Phase Balancing

Considering Customer Load Profile
Priyanka Gangwar, S. N. Singh, and S. Chakrabarti
Department of Electrical Engineering, Indian Institute of Technology Kanpur, Kanpur, India 208016.
Abstract—Unbalanced loading in the distribution systems customer classes is derived in [6] and backtracking search
causes feeder unbalance. This feeder unbalance deteriorates the algorithm is implemented to find the optimal phase sequence
power quality and increases power loss in the system. This paper for each load point. The limitation of the heuristic rule-based
proposes an analytical approach of phase balancing considering
the load profile of each bus. The minimization of power loss methodology is that it re-phases the overall system from the
is taken as an objective, considering the requirement that starting point of the transformer connection, thereby increasing
the number of phase swappings should be minimum. Current the complexity and computational effort increases to achieve
unbalance factor (CUF), voltage unbalance factor (VUF), and the optimal solution. Three-phase balancing is detailed in
loading index improvement are also taken into consideration. A [7] using immune algorithm (IA). A multi-objective function
novel priority vector is also proposed to find the highest-priority
buses, at which phase swapping may be performed. This priority is formed considering customer outage cost, penalty cost,
vector reduces the search space and computational effort of and labour cost; and IA algorithm is used for optimization.
finding the buses at which phase swapping should be carried out. A bacterial foraging-particle swarm optimization (BF-PSO)
The proposed methodology has been tested on IEEE 34-bus and based multi-objective re-phasing strategy is discussed in [8]
IEEE 123-bus system. The result shows significant improvement considering minimization of neutral current, voltage drop,
in current and voltage unbalance factor. Reduction in power loss
is observed, and the loading index of the system is also improved. rephrasing cost, and power loss cost. Fuzzy logic is used
for the objective function formulation in [8]. Reference [9]
Index terms- Load balance, load swapping, phase balancing suggests a phase swapping software based on a graphic
user interface (GUI) using a particular set of rules. Some
meta-heuristic approaches are presented in [10]–[13] using
I. I NTRODUCTION PSO, genetic algorithm (GA), tabu search, and greedy-fuzzy
In the distribution systems, unbalanced loading causes logic, respectively. Most of the presented work considers
current unbalance, which further causes voltage unbalance. only minimization of feeder unbalance. A heuristic method is
Sometimes, minor voltage unbalance at the transformer level detailed in [14] to find the optimal re-phasing solution. In [14],
can create disturbance in the current significantly in all the phase balancing is performed using circuit loss information
loads connected to it. Voltage and current unbalance leads to and current magnitude. Time-varying nature of loads has
increase in power loss, malfunctioning of the relays, and com- also been considered. Dynamic programming based phase
munication interference. Therefore, the operational efficiency balancing algorithm is detailed in [15], and it is compared
and service quality of the distribution system deteriorates with different meta-heuristic methodologies, such as GA, SA,
[1]. Phase balancing can be a solution for the above-stated exhaustive search, backtracking algorithm, and greedy algo-
problems. It can be defined as re-phasing the single or double rithm. Dynamic programming based algorithm shows better
phase loads of a three-phase feeder in order to achieve better performance as compared to other above-stated methodolo-
operational efficiency of the distribution system [2]. gies.
Phase balancing can be performed in two different ways: Most of the above-discussed phase balancing methodologies
one is at the system level by performing network reconfigu- consider only feeder balancing; minimization of power loss,
ration, and the other is at the feeder level by phase swapping. voltage unbalance factor, and improvement in loading index
Network reconfiguration can not efficiently deal with phase are not taken much into consideration. Most of the proposed
balancing problem; it is mainly for load balancing problem methodologies are based on meta-heuristic optimization tech-
[3]. Phase swapping is an efficient way of feeder balancing niques. One of the limitations of meta-heuristic techniques
by swapping the phases of loads which are connected through is that these are iterative techniques based on some initial
the feeder. Phase balancing makes the feeder more balanced in guess. The quality of the optimal solution depends upon the
terms of phases. It improves power quality and reduces power initial guess. However, it is not necessary that the final solution
loss and voltage drop. obtained by these techniques is globally optimal [16]. Compu-
In [4], a mixed-integer programming (MIP) based phase tational time requirement of these iterative algorithms may be
swapping algorithm is presented, where optimal phase pat- more substantial as compared to mathematical programming
tern for minimization of current unbalance is determined. A approaches. This paper proposes an analytical approach of
simulated annealing (SA) based technique is detailed in [5], phase balancing. The main contributions of this paper are as
considering phase swapping cost minimization. The limitation follows.
of SA is that it does not give assurance that the final solution • A novel priority vector is proposed for load swapping
is optimal. Heuristic rule-based phase balancing algorithm is considering load and voltage unbalance factor of each
presented in [6]. Typical load variation curve of different bus, to reduce the search space for phase swapping.
978-1-5386-8218-0/19/$31.00 2019
c IEEE • An algorithm is proposed for phase swapping, consid-
 2

ering varying load profile of the system. The objective It can not be carried out very frequently, unless any exten-
function is minimization of power loss considering the sive overloading or voltage drop conditions arise. It can be
improvement in CUF, VUF, and loading index. performed during the maintenance or restoration period. For a
• The computational time of this methodology is small, given time period, load of each customer for last one month is
even for a large system. taken into account, and the aggregated load is computed for an
Section II of the paper details the phase balancing problem. hour, and a load curve is obtained for that time period. To find
The proposed methodology is discussed in Section III. Section the optimal phase swapping solution for a given time period,
IV presents the implementation of the proposed methodology the discrete load of each customer is taken into account. It
on different test systems. The paper is concluded in Section can be obtained from the load curve. In the case of peak load
V. condition, the phase balancing algorithm considers only peak
load that can also be obtained from the load curve.
II. P ROBLEM F ORMULATION
III. P ROPOSED M ETHODOLOGY
Phase swapping can be done in two ways: one is load
This paper proposes a priority vector consisting of the
swapping at a node, called nodal phase swapping, and the
buses on which load swapping should be performed. The
other is re-taping the lateral branches of the primary feeder
mathematical modeling of the priority vector is given below.
among the phases, that is called lateral phase swapping. If the
• Compute the voltage unbalance factor of each bus con-
lateral phase swapping is performed on a feeder, then on this
sidering the present system conditions. Voltage unbalance
feeder, nodal phase swapping cannot be performed. So, lateral
factor (VUF) of the kth bus is calculated as,
branches can be considered as fictitious bus as on the primary
feeder. Thus, the mathematical formulation of nodal and lateral |V2k |
V U Fk = (1)
phase swapping is same [17]. In this paper, a methodology for |V1k |
nodal phase swapping is proposed.
where |V2k | and |V1k | are magnitude of negative and
positive sequence component of voltage, respectively.
A. Feeder Description • Calculate the total demand of each phase by adding load
Usually, distribution feeder is three-phase, four-wire radial at each bus and each phase and also compute the average
system consisting of loads on single, two, or three phases. demand of the system, savg . Find the overloaded and
Due to unbalance loading on each phase, voltage and current under-loaded phase as follows.
unbalance arises. To reduce the feeder unbalance and system – Overloaded phase- If the demand at the phase is
loss, phase swapping is performed. In this paper, an automatic greater than the average demand of the system.
phase selector switch (PSS) is taken to perform the phase – Under-loaded phase- If the demand at the phase is
swapping of the load point. This PSS is connected between the less than the average demand of the system, and is
primary feeder and each load point. The simplified diagram minimum in all the three phases.
of a distribution feeder with PSS is shown in Fig. 1. • The priority factor considers VUF to account for voltage
unbalance of a bus and load difference of overloaded and
A
under-loaded phase to account for load unbalance. The
B
C
priority factor of the kth bus, pfk , is calculated as,
pfk = V U Fk + λ × (sko − sku ) (2)
PSS PSS

Contro- Contro-
where sko and sku are demand at overloaded and under-
ller ller
loaded phase of kth bus, respectively; λ is a non-negative
Sa Sb Sc Sa Sb Sc
number power of ten. The buses are ranked based on
the priority factor value as obtained in (2). Rank one is
Load Load
assigned to the bus that has the maximum value of pfk .
Fig. 1. Distribution feeder with phase selector switch (PSS)
A. Proposed Algorithm for Nodal Phase Swapping
The input data for phase balancing algorithm are phase
current in each feeder, phase voltage at each bus, and load This paper focuses on nodal phase swapping. In this phase
in each phase and each bus; which is collected from the smart swapping scheme, some loads are swapped between lightly
meters and sent to the supervisory control and data acquisition loaded phase and overloaded phase. The steps are listed below.
(SCADA) system. The output of phase balancing algorithm is Step-1 Calculate the priority factor of each bus and the bus
sent to the SCADA through communication lines, and SCADA that has rank one is the highest priority bus, h, to
sends it to the switch controller of PSS, so that phase swapping perform phase swapping.
can be carried out [18]. Step-2 Compute the current deviation in the incoming feeder
that is connected to the bus, h, as given below.
B. Load Curve Considering Customer Load Profile xh = Iaj − Iavg
j

The optimal solution of the phase balancing algorithm yh = Ibj − Iavg

j
(3)
depends on the time-varying nature of customer load profile. zh = Icj − j
Iavg
 3

where Iaj , Ibj , and Icj are the currents in incoming Load swapping is allowed only when it results
•
j
feeder j of phase a, b, and c, respectively; Iavg is in a significant amount of reduction in power
the average current in the incoming feeder j. loss, so that the number of swapping can be
Step-3 Find the minimum and maximum value among xh , minimized.
yh , and zh . Let, xh be the maximum and yh be the Step-6 When the under-loaded phase gets loaded to more
minimum. Then, load of hth bus, phase a, is swapped than average loading of the system, stop the load
with phase b. This means, load swapping takes swapping. The flow chart of the proposed methodol-
place between the phases that have maximum and ogy is shown in Fig. 2.
minimum current deviation with respect to average
current.
Step-4 Perform the load flow after load swapping. The Start
forward-backward sweep methodology is used to
perform power flow for the unbalanced distribution Read the load data, current, voltage data
and system topology
system [19]. Calculate the power loss and current
unbalance factor (CUF) of the primary feeder after Find the highest priority bus (using 2) at
load swapping. The total power loss, PT L , is com- which load swapping should be performed
puted as,
N
X −1 Find the phases in between load swapping
PT L = Pl(m,m+1) (4) have to perform as discussed in Step 2-3
m=1
Perform the load swapping and find the
where N is the total number of buses in the system; losses and CUF using (5)-(6)
Pl(m,m+1) is the power loss in the line between bus
m and m+1. Pl is estimated as,
Losses and
∗ ∗
 
No CUF are reduced after
Pl(m,m+1)=Re V mI (m,m+1)−V m+1I (m,m+1) (5)
load
swapping
where V m is the vector of three-phase voltage at mth
∗
bus; I (m,m+1) is the vector of three-phase current Yes
between bus m and m+1. The CUF of the primary
feeder is computed as, Total demand
of under-loaded
No
|I2j | phase is greater than
CU Fj = (6) average load of
|I1j | the system
where |I2j | and |I1j | are magnitude of negative and
Yes
positive sequence current component of the jth
feeder. Optimal re-phasing strategy
Step-5 Load swapping is considered, if PT L and CU F is
reduced after load swapping; otherwise load swap- Stop
ping is performed on the next highest priority node.
The following constraints have also been taken into Fig. 2. Proposed methodology of phase swapping
consideration.
IV. C ASE S TUDY
• The voltage magnitude at the kth bus, phase p,
should be within the pre-specified limit. The proposed algorithm is tested on IEEE 34-bus and IEEE
123-bus distribution test feeders. All the line and load data are
Vkmin ≤ Vkp ≤ Vkmax (7) taken from [21]. The aggregate of the hourly load profile is
• The current in the feeder between the buses m taken into consideration to perform the phase balancing. The
and m+1, phase p, should be within the pre- single line diagram of IEEE 34-bus and IEEE 123-bus systems
specified limit. are shown in Figs. 3 and 4.
min p max
I(m,m+1) ≤ I(m,m+1) ≤ I(m,m+1) (8) 848

822 846

• Load balancing index of the system, Ls , should 820 844

not deteriorate after load swapping. It can be 818
864
842
computed as [20], 802 806 808 812 814 850 824 826
858
834 860 836
840
816
N −1
1 X sm 800
832
888 890
862

Ls = (9) 810
N − 1 m=1 smax
m 852
838

where sm is power flow in the branch m; smax

m 828 830 854 856

is the maximum capacity of the branch m.

Fig. 3. IEEE 34-bus distribution system
 4

32 29 250 350
33
30
251 51 111 110 112 113 114 day for 24 hours. The average number of swapping is 3 from
28 50 151 300
31 49
25
48
47
46
109 107 one day to another day to perform load balancing. The runtime
26 108 451
45 104
27
23 44 43 65
64 106
103
450
of this algorithm is 1 second for IEEE 34-bus system and 2
105 102
24 42 41
66
63
101
100
seconds for IEEE 123-bus system in a core i5, 2.9 GHz, 4GB
21 99 71
22 40
38
39 62 197
98
70 RAM computer.
19 35 36 97 69
20 18 135 68 75
37 160 67
60 74
57 73 TABLE II
14 58
11 59
61
72
79 85 O PTIMAL R E - PHASING S TRATEGY
610
10
9 53 54 77 78
2 52 56
152 55
7 8 13 76
80
Bus number From phase To phase
94 84
149 1 34 96 76
88
IEEE 34-bus system
12 90 81
150 17 92
862 b a
15 87 86 83
3 5
95
93 91 89 82 844 b a
6 16 195
4 IEEE 123-bus system
98 a b
Fig. 4. IEEE 123-bus distribution system 82 a b
To calculate the priority vector, load at each bus is taken 29 a b
in per unit and lambda is taken as 10. The priority vector
for IEEE 34-bus system, pv34 , and for IEEE 123-bus system,
pv123 , are given as, TABLE III
C OMPARISON WITH D IFFERENT M ETHODOLOGIES FOR IEEE 34- BUS
S YSTEM
pv123 = [64 98 82 99 79 63 80 77 87 53]
Cases GA Tabu search Proposed
pv34 = [862 844 824 846 834 836 848 860 842 840] Reduction in loss (%) 7.15 7.52 9
Improvement in CUF (%) 45 46 54
Priority vector has the highest priority buses for load swapping Improvement in load
12.5 11.1 14.1
based on their priority factor value as calculated from (2). balancing index (%)
Computational time (secs) 42 11 1
TABLE I
S YSTEM P ERFORMANCE OF IEEE 34- BUS S YSTEM AND IEEE 123- BUS
S YSTEM A FTER P HASE BALANCING 100
Before phase balancing
IEEE 34-bus system 80 After phase balancing
Cases
Before phase After phase 60
CUF (%)

balancing balancing
Power loss (kW) 686.5 623.1 40

CUF of primary feeder (%) 42.8 19.6 20

Load balancing index 0.0452 0.0388
0
Minimim voltage (pu) 0.93 0.94 5 10 15 20 25 30
Branch number
Number of phase swapping - 2
Computational time (seconds) - 1
Fig. 5. Improvement in CUF of IEEE 34-bus system
IEEE 123-bus system
Power loss (kW) 1815.2 1745.2
4
CUF of primary feeder (%) 25.66 17.67 Before phase balancing
Load balancing index 0.4394 0.4215 After phase balancing
3
Minimim voltage (pu) 0.89 0.91
VUF (%)

Number of phase swapping - 3 2

Computational time (seconds) - 2
1

In IEEE 34-bus system, phase b is overloaded and phase a is

0
under-loaded as compared to phase c. Bus 862 is a three-phase 1 11 21 31 41 51 61
Bus number
71 81 91 101 111 121

bus consisting of load at phase b only, and bus 844 is also a

three-phase bus consisting of load at phase b and c. The load Fig. 6. Improvement in VUF of IEEE 123-bus system
at bus 862 and 844, phase b, are shifted to phase a to perform In IEEE 123-bus system, phase a is overloaded, and phase b
the phase balancing operation. Only two times load swapping is under-loaded as compared to phase c. Buses 98, 82, and 29
has been performed, and significant improvement can be seen are three-phase buses, and consist of load only at phase a. To
in the system performance, as presented in Table I. After make the phase more balanced, load of these buses are shifted
phase balancing, power loss and current unbalance factor of to phase b from phase a. Only three-times load swapping is
the primary feeder is reduced by 9% and 54%, approximately. formed, and system performance is improved significantly as
Load balancing index and the minimum voltage of the system presented in Table I. Figs. 6 and 7 show that the VUF and CUF
is also improved. Fig. 5 shows that the current unbalance factor of most of the buses and feeders are improved significantly
of each branch is reduced significantly after phase balancing. after phase balancing. For the single-phase buses and feeders,
Fig. 9 shows the reduction in losses after the load balancing VUF and CUF are taken as zero. The optimal nodal phase
throughout the year considering the average load profile for a balancing strategy for both the systems are presented in Table
 5

60
Before phase balancing R EFERENCES
50 After phase balancing
[1] T.-H. Chen and J.-T. Cherng, “Optimal phase arrangement of distribu-
40
tion transformers connected to a primary feeder for system unbalance
CUF (%)

30 improvement and loss reduction using a genetic algorithm,” IEEE

Transactions on Power Systems, vol. 15, no. 3, pp. 994–1000, Aug 2000.
20
[2] M. Terorde, H. Wattar, and D. Schulz, “Phase balancing for aircraft
10 electrical distribution systems,” IEEE Transactions on Aerospace and
0 Electronic Systems, vol. 51, no. 3, pp. 1781–1792, July 2015.
20 40 60 80 100 120
Branch number
[3] Y. . Hsu, J. . Yi, S. S. Liu, Y. W. Chen, H. C. Feng, and Y. M.
Lee, “Transformer and feeder load balancing using a heuristic search
Fig. 7. Improvement in CUF of IEEE 123-bus system approach,” IEEE Transactions on Power Systems, vol. 8, no. 1, pp. 184–
190, Feb 1993.
[4] H. M. Khodr, I. J. Zerpa, P. M. D. O. de Jesu’s, and M. A. Matos,
II. The proposed methodology is compared with the existing “Optimal phase balancing in distribution system using mixed-integer
linear programming,” in 2006 IEEE/PES Transmission Distribution
methodologies [11] and [12]. Table III shows that the proposed Conference and Exposition: Latin America, Aug 2006, pp. 1–5.
methodology performs better as compared to the existing [5] J. Zhu, G. Bilbro, and M.-Y. Chow, “Phase balancing using simulated
methods for both the systems. annealing,” IEEE Transactions on Power Systems, vol. 14, no. 4, pp.
1508–1513, Nov 1999.
[6] C.-H. Lin, C.-S. Chen, H.-J. Chuang, and C.-Y. Ho, “Heuristic rule-
based phase balancing of distribution systems by considering customer
Phase a
Phase b
load patterns,” IEEE Transactions on Power Systems, vol. 20, no. 2, pp.
120 Phase c 709–716, May 2005.
[7] M. . Huang, C. . Chen, C. . Lin, M. . Kang, H. . Chuang, and C. .
Load (kW)

Huang, “Three-phase balancing of distribution feeders using immune

80 algorithm,” IET Generation, Transmission Distribution, vol. 2, no. 3,
pp. 383–392, May 2008.
[8] R. A. Hooshmand and S. Soltani, “Fuzzy optimal phase balancing of
40 radial and meshed distribution networks using bf-pso algorithm,” IEEE
0 1000 2000 3000 4000 5000 6000 7000 8000 Transactions on Power Systems, vol. 27, no. 1, pp. 47–57, Feb 2012.
Hours
[9] C.-C. Kuo and Y.-T. Chao, “Energy management based on am/fm/gis for
phase balancing application on distribution systems,” Energy Conversion
Fig. 8. Load profile of bus no. 890 of IEEE 34-bus system for the whole
and Management, vol. 51, no. 3, pp. 485–492, 2010.
year
[10] B. M. Kalesar, “Customers swapping between phases for loss reduction
considering daily load profile model in smart grid,” 2016.
[11] H. Jiang, Y. Sun, S. Gao, and T. Li, “Loss reduction strategy of
0.5 low-voltage distribution network based on phase swapping,” in IOP
Conference Series: Earth and Environmental Science, vol. 170, no. 4.
0.4 IOP Publishing, 2018, p. 042040.
Reduction in
losses (pu)

[12] M. Lafortune, D. Bouchard, and J. Morelli, “Phase swapping for distri-

0.3
bution system using tabu search,” in WSEAS International Conference
0.2 on Energy Planning, Energy Saving, Environmental Education, 2007,
pp. 67–71.
0.1 [13] M. Sathiskumar and S. Thiruvenkadam, “Phase balancing of unbalanced
50 100 150 200 250 300 350
Day distribution network through hybrid greedy-fuzzy algorithm,” Interna-
tional Journal of Computer Applications, vol. 34, no. 6, pp. 38–45,
Fig. 9. Reduction in losses in a year after load balancing considering 2011.
aggregated load profile for a day for IEEE 34-bus system [14] M. Dilek, R. P. Broadwater, J. C. Thompson, and R. Seqiun, “Simul-
taneous phase balancing at substations and switches with time-varying
load patterns,” IEEE Transactions on Power Systems, vol. 16, no. 4, pp.
V. C ONCLUSION 922–928, Nov 2001.
[15] K. Wang, S. Skiena, and T. G. Robertazzi, “Phase balancing algorithms,”
This paper proposed a methodology of nodal phase balanc- Electric Power Systems Research, vol. 96, pp. 218–224, 2013.
[16] M. Madić, D. Marković, and M. Radovanović, “Comparison of meta-
ing to make the feeder more balanced. The load is swapped heuristic algorithms for solving machining optimization problems,”
between overloaded phase and under-loaded phase. A novel Facta universitatis-series: Mechanical Engineering, vol. 11, no. 1, pp.
priority vector is also proposed to find the buses at which 29–44, 2013.
[17] J. Zhu, M.-Y. Chow, and F. Zhang, “Phase balancing using mixed-integer
nodal load swapping should be carried out. This priority programming [distribution feeders],” IEEE Transactions on Power Sys-
vector reduces the search space and computation time of the tems, vol. 13, no. 4, pp. 1487–1492, Nov 1998.
algorithm. As a result, the execution time of this algorithm [18] S. Soltani, M. Rashidinejad, and A. Abdollahi, “Stochastic multiobjec-
tive distribution systems phase balancing considering distributed energy
is small even for large systems. It is implemented on the resources,” IEEE Systems Journal, vol. 12, no. 3, pp. 2866–2877, Sept
IEEE 34-bus system and IEEE 123-bus system, and the overall 2018.
system performance improved significantly. Power loss, CUF, [19] P. Gangwar, S. N. Singh, and S. Chakrabarti, “Network reconfiguration
for unbalanced distribution systems,” in TENCON 2017 - 2017 IEEE
and VUF are reduced after phase balancing. Region 10 Conference, Nov 2017, pp. 3028–3032.
[20] M. A. Kashem, V. Ganapathy, and G. B. Jasmon, “Network reconfigu-
ration for load balancing in distribution networks,” IEE Proceedings -
Generation, Transmission and Distribution, vol. 146, no. 6, pp. 563–567,
VI. ACKNOWLEDGEMENT Nov 1999.
[21] D. T. Feeders, “Ieee pes distribution system analysis subcommittee,”
The authors would like to thank Ministry of Power, Gov- OnlineAvailable: http://sites.ieee.org/pes-testfeeders/resources/, 2011.
ernment of India, and IIT Kanpur for providing financial
support to carry out this research work under project no.
MP&IIT/EE/2014297.