Electrical Power and Energy Systems 63 (2014) 609–617

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

A MINLP technique for optimal placement of multiple DG units

in distribution systems
Sandeep Kaur ⇑, Ganesh Kumbhar, Jaydev Sharma
Department of Electrical Engineering, IIT Roorkee, Roorkee 247667, India

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

Article history: Loss reduction is one of the prime objectives for planning of distributed generation (DG). To improve per-
Received 11 September 2013 formance of distribution systems, optimal placement of distributed generator is critically important, as
Received in revised form 30 May 2014 DG benefits are site and size specific. Optimal placement of multiple DG units is non-convex and non-
Accepted 13 June 2014
linear optimization problem. The methodology proposed in this paper solves Mixed Integer Non-Linear
Available online 12 July 2014
Programming (MINLP) formulation for loss minimization. The proposed methodology simplifies the
problem by dividing it in two phases, namely Siting Planning Model (SPM) and Capacity Planning Model
Keywords:
(CPM) thereby reducing the search space and computational time. The SPM model selects the candidate
Distributed generation (DG)
Mixed Integer Nonlinear Programming
buses based on Combined Loss Sensitivity (CLS). In CPM, optimal locations and sizes are obtained by
(MINLP) integrating Sequential Quadratic Programming (SQP) and Branch and Bound (BAB) algorithm for MINLP
Optimal DG placement problem. The performance of the proposed method is tested on IEEE 33-bus and IEEE 69-bus distribution
Branch and Bound (BAB) system for placement of single and multiple DG units capable of delivering either real or both real and
Sequential Quadratic Programming (SQP) reactive power. The obtained results are then compared with three basic classes of optimization methods,
viz., Exhaustive Load Flow (ELF), Improved Analytical (IA) and Particle Swarm Optimization (PSO). The
key advantage of the proposed method is that it simultaneously considers placement of multiple DG units
giving better optimal solution in reasonably less computational time.
Ó 2014 Elsevier Ltd. All rights reserved.

Introduction and siting of DG can be broadly classified in three categories: (i)

analytical methods [3–9], (ii) numerical methods (Linear Program-
In the past, the distribution system was treated as a passive ming (LP), Quadratic Programming (QP), Non-Linear Programming
element of the power system. In the recent years due to penetra- (NLP), Mixed Integer Non-Linear Programming (MINLP), Kalman
tion of distributed generation, it has become an active network. Filter Algorithm, etc.) [10–19], and (iii) intelligent search based
In addition, distributed generation has emerged as an immediate methods (Genetic Algorithm (GA), Particle Swarm Optimization
rescue to address environmental issues, meet increasing demand (PSO), Ant Colony (AC), Artificial Bee Colony (ABC), Differential
and defer investment on new transmission and distribution Evolution, etc.) [20–31].
network. In [3], ‘2/3 rule’ (widely used for capacitor placement) is used
Most of the distribution networks are either radial or weakly for DG placement. However, this method is based on the assump-
meshed in nature. Due to high R/X ratio, distribution systems cause tion of uniform line loading, and cannot be used for non-uniform
a large voltage drop and substantial power losses. Distribution sys- loading and meshed systems. Wang and Nehrir [4] derived analyt-
tem contributes a major share of total power losses in whole power ical expressions to determine the optimal location of DG in the
system. DG resources not only reduce line losses, but also improve radial as well as meshed systems for minimizing power losses.
the voltage profile. However, studies have indicated that the However, the authors did not consider DG size as a variable.
improper placement of DG units might lead to increased line Abu-Mouti and El-Hawary [5] proposed best location by offsetting
losses. each load. In [6,7], analytical expressions are derived to calculate
Various techniques of optimal placement of DG units have been the optimum size of the single DG unit with unity power factor.
nicely reviewed in [1,2]. The solution algorithms for optimal sizing Hung [8] proposed another analytical expression for optimal sizing
and power factor for four types of DG units. However, these formu-
⇑ Corresponding author. Tel.: +91 9758457273; fax: +91 1332273560. lations are limited to placement of single DG with both real and
E-mail addresses: sandipsaroa@gmail.com (S. Kaur), gkumbfee@iitr.ac.in reactive power capability. The placement of multiple DG units is
(G. Kumbhar), jaydsfee@gmail.com (J. Sharma). addressed in [9]. Since this method considers the sequential

http://dx.doi.org/10.1016/j.ijepes.2014.06.023
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
610 S. Kaur et al. / Electrical Power and Energy Systems 63 (2014) 609–617

Nomenclature

Vi, Vj voltage at ith and jth bus |Vi|min, |Vi|max minimum and maximum permissible bus
Gij, Bij conductance and susceptance of branch con- voltage
necting ith bus to jth bus ðPDG Þmin ; ðPDG Þmax minimum and maximum real power injection
Rij, Xij resistance and reactance of branch connect- by DG unit
ing ith bus to jth bus ðQ DG Þmin ; ðQ DG Þmax minimum and maximum reactive power
Pi, Qi real and reactive power injected in ith bus injection by DG unit
Ploss, Qloss, Sloss real reactive and total power loss in distribu- ðPiDG Þ real power injected by DG unit at ith bus
tion system
ðQ iDG Þ reactive power injected by DG unit at ith bus
NBr number of branches
N number of buses Nmax
DG maximum number of DG units on candidate
di, dj voltage angle at ith, jth bus buses in distribution system
PiG real power generation at ith bus rkDG decision variable for DG unit placement at kth
PiD real power demand at ith bus candidate bus
Q iG reactive power generation at ith bus pfDGlower ; pfDGupper lower and upper limit for DG power factor
Q iD reactive power demand at ith bus Sij(max) power flow (MVA) limit of feeder section be-
PGðmaxÞ maximum real power generation at slack bus tween ith and jth bus
Q GðminÞ ; Q GðmaxÞ minimum and maximum reactive power gen- Sij power flow (MVA) in feeder section between
eration at slack bus ith and jth bus

placement of multiple DG units, sometimes it might lead to burden, an approach based on integrated Sequential Quadratic Pro-
sub-optimal solution. gramming (SQP) and Branch and Bound (BAB) is adopted. The per-
To allocate the distributed generation optimally, many numer- formance of the proposed formulation is compared with the results
ical techniques have also been employed. In [10], the classical Kal- of the recently published work [9,27]. To the best of the author’s
man Filter Algorithm is used to compute DG size and location. knowledge, the proposed model is the only integrated model with
However, the locations obtained based on maximum and mini- MINLP formulation for loss minimization.
mum load in the specific area may not be optimal. El-Khattam The organization of the paper is as follows. The detailed
et al. [11] proposed Mixed Integer Non-Linear Programming (MIN- modelling of SPM and CPM models is discussed in Section ‘Problem
LP) based DG planning model for cost minimization. An LP based formulation’. A solution algorithm integrating SQP and BAB is pre-
model for maximizing DG injection in deregulated environment sented in Section ‘Solution algorithm’. The proposed model is
is described in [12]. In [13], a multi-objective MINLP model for a implemented on 33-bus and 69-bus system. Simulation results
tradeoff between operating cost and DG energy is proposed for are discussed in Section ‘Results and discussion’, followed by
rural microgrid. Rueda-Medina et al. [14] formulated optimal siz- conclusions in Section ‘Conclusion’.
ing and siting problem as Mixed Integer Linear Programming
(MILP) problem. Since the optimal placement DG unit is MINLP for- Problem formulation
mulation, linear model might affect the power flows in some cases.
Energy loss minimization for stochastic renewable generation is Optimal allocation of DG is non-convex mixed integer nonlinear
formulated in [15,16]. DG allocation for reliability improvement problem. Due to the inherent nonlinearity and exhaustive search
and line loss minimization is addressed in [17]. A novel approach space, these formulations become computationally extensive and
for optimal allocation of CHP based DG for integrated energy model sometimes fail to converge to the optimal solution. To reduce this
of gas, water, and electricity and its operational performance is computational burden, this optimization problem is solved as
described in [18]. two-tier model, namely Siting Planning Model (SPM) and Capacity
The optimization problem of DG placements is also addressed Planning Model (CPM).
using many heuristic techniques such as Genetic Algorithm [19–
22], Particle Swarm [23–28], Ant Colony [29], Artificial Bee Colony Siting Planning Model (SPM)
[30], and Differential Evolution [31]. Although, the heuristic algo-
rithms are derivative free and simple to implement, they need sev- The objective of SPM is to determine the list of prospective best
eral iterations to ensure converged solution. Thus, they become locations for DG units. In this model, the loss sensitivity of each bus
computationally intensive. In addition, search based methods to with respect to real, reactive and apparent power injection is
some extent depend on tuning parameters. If tuning parameters calculated as (1) [32,33].
are not carefully chosen, these algorithms may lead to a sub-opti- 9
@Sloss
¼ @P@Plossi þ j @Q@Ploss =
mal solution. Even with carefully chosen tuning parameters, not @P i i
ð1Þ
every execution of the algorithm may lead to the same solution. @Sloss
@Q i
¼ @P@Qlossi þ j @Q@Qlossi ;
On the other hand, the appropriate numerical method with suit-
able initial conditions leads to consistent optimal or nearly optimal The Combined Loss Sensitivity (CLS) at each bus is calculated as
solution. (2) [33].
In this paper, MINLP based formulation is proposed and solved  
 @Ploss @Q loss 
in two stages, viz., Siting Planning Model (SPM) and Capacity Plan-  @P @P i 
CLS ¼  @P i 
@Q loss 
ð2Þ
ning Model (CPM). Firstly, SPM identifies the potential candidate  @Qlossi @Q i 
buses through sensitivity analysis to reduce the search space. A list
of top ranked potential buses with higher sensitivity is passed to The buses are then arranged in descending order of their Com-
CPM. In CPM, MINLP based formulation is solved to select final bined Loss Sensitivity values. To ensure the optimal solution, a list
optimal locations and DG sizes. To reduce the computational of the top 30% buses is prepared as potential candidates for CPM.
 S. Kaur et al. / Electrical Power and Energy Systems 63 (2014) 609–617 611

Search space is also reduced significantly, since a small percentage Solution algorithm
of total number of buses is passed as potential candidates to the
CPM model. Optimization problem (3–13) is modeled as follows.
9
Min: f ðx; yÞ >
Capacity Planning Model (CPM) >
>
s:t gðx; yÞ 6 0 =
ð14Þ
In CPM, MINLP based formulation for optimal locations (out of s:t hðx; yÞ ¼ 0 >
>
>
;
the potential candidate buses from SPM) and size is defined by fol- x 2 X; y 2 f0; 1gk

lowing objective function and constraints.

Objective function to minimize the real power losses in the dis- where x is set of continuous variables and y is the set of binary deci-
tribution system is given as sion variables for locations of DG unit at k candidate buses.
In the proposed work, an approach based on the integration of
X
NBr
Sequential Quadratic Programming (SQP) and Branch and Bound
F¼ Gij ½ðV i Þ2 þ ðV j Þ2  2ðV i ÞðV j Þ cosðdi  dj Þ where
ij¼1
(BAB) techniques is adopted to solve MINLP formulation. This algo-
rithm possesses better computational performance and strong con-
i; j 2 f1; 2 . . . Ng and ij 2 f1; 2 . . . NBrg ð3Þ
vergence property [34]. The steps of the algorithm are as follows
Following are the constraints of this optimization problem. and the flow chart is shown in Fig. 1.

(I) Power balance: sum of all incoming and outgoing real and I. Set upper bound U = 1. All binary variables yi (i = 1, 2, 3, . . .)
reactive power at each bus must be zero. It can be repre- are relaxed and resulting NLP problem is solved using SQP
sented as (4) and (5). algorithm with following formulation (15).
9
T k kT k
X
N
min f ðkÞ þ rf ðkÞ d þ 12 d wðkÞ dwhere wðkÞ is hessian of >
>
>
PiG  PiD  jV i jjV j jðGij cosðdi  dj Þ þ Bij sinðdi  dj ÞÞ >
>
>
i¼1
Langrangian function given as >
=
P
¼ 0 where i 2 f1; 2 . . . Ng and ij 2 f1; 2 . . . NBrg ð4Þ LðX; YÞ ¼ f ðX; YÞ þ ki gðX; YÞ þ li hðX; YÞ
>
>
T k
s:t gk þ rg k d 6 0 >
>
>
>
X
N k k
>
;
k k
x þ dx 2 X; y þ dy 2 Y
Q iG  Q iD  jV i jjV j jðGij sinðdi  dj Þ  Bij cosðdi  dj ÞÞ ¼ 0 ð5Þ
i¼1 ð15Þ
(II) System voltage: voltage magnitude of all the buses must be II. If variables y are 0 or 1, the solution is obtained. Otherwise,
within the allowed upper and lower limits. go to the next step.
jV i jmin 6 jV i j 6 jV i jmax ð6Þ III. Select a variable yi, which does not have value 0 or 1. Gener-
ate a binary tree with edges from root node as 0 and 1 indi-
(III) Distribution substation capacity: the slack bus power must be cating the values of the selected variable to the NLP
within permissible limit of substation capacity. relaxation. Select one of the two relaxed NLP problems and
0 6 PiG 6 PGðmaxÞ i 2 fSlack busg ð7Þ solve (15). If the problem is infeasible, then fathom the node
and go to other unexplored node, otherwise proceed to the
next step.
Q GðminÞ 6 Q iG 6 Q GðmaxÞ ð8Þ IV. Update the solution vector as (16)

(IV) DG capacity limit at candidate bus: the generated power of DG ðxkþ1 ; ykþ1 Þ
k
ðxk ; yk Þ þ ðdx ; dy Þ
k
ð16Þ
units must lie within minimum and maximum pre-specified
k+1
DG capacity. V. If the solution is optimal and y is 0 or 1, then update cur-
ðPDG Þmin rkDG 6 PkDG 6 ðP DG Þmax rkDG k rent best as (17), (18) and go for next unexplored node.
ðx ; y Þ ðxkþ1 ; ykþ1 Þ ð17Þ
2 fSet of candidate DG busesg ð9Þ
f f ðxkþ1 ; ykþ1 Þ and U f ð18Þ
ðQ DG Þmin rkDG 6 Q kDG 6 ðQ DG Þmax rkDG ð10Þ k+1
VI. If y is non-integer, generate two sub-problems with bin-
(V) Number of DG units: total number of DG units installed on ary variable as 0 and 1 and solve using (15).
candidate buses should not exceed the maximum permissi- VII. If a solution is not optimal then calculate integrity gap (d) as
ble DG units to be installed. (19).
0 6 NDG 6 Nmax
DG N DG 2 fNo: of DG units placed on candidate busesg
d maxjykþ1
i  roundðyikþ1 Þj ð19Þ
ð11Þ
VIII. If (d < s) then examine other unexplored node.
(VI) Power factor of DG units: power factor of DG should adhere to IX. If (d > s) and binary variable appear to converge to fractional
the upper and lower DG power factor limits. integer value, go to step (III).
PiDG X. If all the nodes are examined end the program, else go for
pfDGlower < qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi < pfDGupper ð12Þ other unexplored node.
2 2
ðPiDG Þ þ ðQ iDG Þ
Results and discussion
(VII) Line power flow: maximum power flow in a feeder section
should not exceed the thermal limit of the feeder section. In this section, a number of case studies are presented to
The line flow limit can be expressed as (13) show the effectiveness of the proposed MINLP Algorithm. The
0 6 Sij 6 SijðmaxÞ where ij 2 f1; 2; . . . NBrg ð13Þ SPM and CPM models are implemented in MATLAB and AMPL
612 S. Kaur et al. / Electrical Power and Energy Systems 63 (2014) 609–617

Fig. 1. Flow chart for MINLP based algorithm.

(A Mathematical Programming Language), respectively. Knitro Placement of DG units in IEEE 33 bus system
solver [35] was used to solve the proposed MINLP formulation on
personal computer with 2.93 GHz, Intel core™ 2 duo CPU with The total real and reactive power demands in IEEE 33-bus sys-
4 GB RAM using windows XP. The formulation is tested on IEEE tem are 3.7 MW and 2.3 MVAr respectively. System line loss with-
33 bus and 69 bus distribution systems. The results are then com- out DG is 211 kW. The list of potential buses for IEEE 33-bus
pared with the Exhaustive Load Flow (ELF), Improved Analytical system provided by SPM model is {5, 6, 8, 9, 10, 13, 24, 28, 29,
(IA) [8,9] and Particle Swarm Optimization (PSO) [27], which are 30}. The proposed MINLP based CPM model is solved with reduced
the promising methods for multiple DG placements. search space to obtain optimal DG locations and sizes.
All the simulations are carried out considering the system peak The simulation results of placement of single and multiple DG
load. The lower and upper limits of bus voltages used in the simu- units with real power are presented in Table 1. All the methods
lations are 0.90 pu and 1.05 pu respectively [9]. In addition, the converge to the same solution for single DG placement. However,
maximum DG power is constrained such that it should not exceed MINLP based proposed formulation takes least time to converge.
the total load and losses. The results are obtained with DG units of The major advantage of the proposed formulation is evident from
Type-1 (capable of delivering real power) and Type-3 (capable of placement of multiple DG units. In the case of 2 DG units, mini-
delivering real and reactive power). However, proposed formula- mum losses and with lowest DG size is obtained with the proposed
tion is generalized and can be implemented for any type of DGs. method in comparison to the other methods. The proposed loca-
The power factor of DG unit, capable of providing real and reactive tions are found to be optimal as verified by exhaustive search for
power, is constrained between 0.8 and 1.0. 2 and 3 DG scheme. Moreover, in the case of 3 DG units, proposed
 S. Kaur et al. / Electrical Power and Energy Systems 63 (2014) 609–617 613

Table 1
Placement of DG units with real power capability for IEEE 33-bus system.

No. of DGs Method Bus no. DG power (MW) Losses (kW) Loss reduction (%) Time (s)
1 DG ELF 6 2.60 111.10 47.39 1.06
IA 6 2.60 111.10 47.39 0.16
MINLP 6 2.59 111.01 47.39 0.09
PSO 6 2.59 111.10 47.39 –
2 DG ELF 12 1.02 87.63 58.51 2.03
30 1.02
IA 6 1.80 91.63 56.61 0.27
14 0.72
MINLP 13 0.85 87.16 58.69 0.80
30 1.15
PSO 12 1.00 87.50 58.52 –
30 1.02
3 DG ELF 13 0.90 74.27 64.83 3.06
24 0.90
30 0.90
IA 6 0.90 81.05 61.62 0.40
12 0.90
31 0.72
MINLP 13 0.80 72.79 65.50 1.20
24 1.09
30 1.05
PSO 13 0.88 73.20 65.34 –
24 1.09
30 1.01

Table 2
Placement of 1 DG unit with real and reactive power capability for IEEE 33-bus system.

Method Bus no. DG power Total DG power Optimal power factor Losses (kW) Loss reduction (%)
MW MVAR MVA
IA 6 2.637 1.634 3.102 0.850 68.157 67.69
MINLP 6 2.558 1.761 3.105 0.823 67.854 67.84
PSO 6 2.557 1.746 3.096 0.826 67.857 67.84

Table 3
Placement of 2 DG units with real and reactive power capability for IEEE 33-bus system.

Method Bus no. DG power Total DG power Optimal power factor Losses (kW) Loss reduction (%)
MW MVAR MVA
IA 6 1.800 1.115 3.177 0.850 44.84 78.77
30 0.900 0.557
MINLP 13 0.819 0.434 2.477 0.883 29.31 86.10
30 1.550 1.240 0.800
PSO 12 0.818 0.566 3.774 0.822 39.10 81.49
29 1.699 1.191 0.819

Table 4
Placement of 3 DG units with real and reactive power capability for IEEE 33-bus system.

Method Bus no. DG power Total DG power Optimal power factor Losses (kW) Loss reduction (%)
MW MVAR MVA
IA 06 0.900 0.557 2.859 0.85 23.05 89.09
14 0.630 0.390
30 0.900 0.557
MINLP 13 0.766 0.411 3.481 0.87 12.74 93.96
24 1.044 0.552 0.88
30 1.146 0.859 0.80
PSO 13 0.764 0.535 3.395 0.82 15.0 92.89
24 1.068 0.613 0.87
30 1.016 0.691 0.83
614 S. Kaur et al. / Electrical Power and Energy Systems 63 (2014) 609–617

1.02

0.98
Bus voltage (pu)

0.96

0.94

0.92 Without DG
2 DG at bus 6 & 30 (IA)
2 DG at bus 12 and 29(PSO)
2 DG at bus 13 & 30 (MINLP)
0.9
0 5 10 15 20 25 30 35
Bus number

Fig. 2. Voltage profile of IEEE 33-bus system with 2 DG units.

1.02

1
Bus voltage (pu)

0.98

0.96

0.94

Without DG
0.92 3 DG at bus 6,14 & 30 (IA)
3 DG at bus 13, 24 & 30 (PSO)
3 DG at bus 13, 24 & 30 (MINLP)
0.9
0 5 10 15 20 25 30 35
Bus number

Fig. 3. Voltage profile of IEEE 33-bus system with 3 DG units.

Table 5
Critical voltages on IEEE 33 bus system with DG.

Case Method Minimum voltage pu (Bus no) Maximum voltage pu (Bus no)
UPF NUPF UPF NUPF
Without DG 0.904(18) 1.0(1)
1 DG IA 0.9425(18) 0.9575(18) 1.0000(1) 1.0007(6)
MINLP 0.9424(18) 0.9584(18) 1.0000(1) 1.0010(6)
PSO 0.9424(18) 0.9598(18) 1.0000(1) 1.0029(6)
2 DG IA 0.9539(33) 0.9600(18) 1.0000(1) 1.0031(6)
MINLP 0.9685(33) 0.9804(25) 1.0000(1) 1.0010(13, 30)
PSO 0.9650(18) 0.9828(25) 1.0000(1) 1.0178(29)
3 DG IA 0.9690(18) 0.9821(25) 1.0000(1) 1.0006(14)
MINLP 0.9687(18, 33) 0.9924(8) 1.0000(1) 1.0010(30)
PSO 0.9684(33) 0.9892(33) 1.0000(1) 1.0020(13)

method gives minimum losses with slightly higher DG size than IA Since Exhaustive Load Flow (ELF) explores entire search space,
method. it gives most accurate results. Although, ELF is straightforward
Tables 2–4 presents the simulation results of placement of sin- and simple, it is very time consuming. In IA method, multiple DG
gle and multiple DG units with real and reactive power capability. units are placed sequentially, i.e. next optimal location is obtained
In Table 2, MINLP based proposed formulation shows its improved in the presence of previously placed DG units. Even though, IA
performance in reaching most optimum solution. In the case of 2 method is quite fast, the separate evaluation of multiple DG units
DG units (Table 3), it can be observed that the proposed method may lead to sub-optimal locations [1]. As observed in Tables 3
reaches to the optimal locations and sizes, thereby giving and 4, IA method leads to sub-optimal locations for multiple DG
maximum loss reduction with least total DG power. In the case units. For example, to place 2 and 3 DG units, optimal locations
of 3 DG units (Table 4), optimal locations are identical for MINLP given by IA method are bus numbers {6, 30} and {6, 14, 30}, respec-
and PSO. However, proposed method gives maximum loss reduc- tively. However, the actual optimal locations, to place 2 and 3 DG
tion with DG size slightly higher than PSO. units, are bus numbers {13, 30} and {13, 24, 30}, respectively.
 S. Kaur et al. / Electrical Power and Energy Systems 63 (2014) 609–617 615

Table 6 maintaining the balance between explorative and exploitative

Placement of DG units with real power capability for IEEE 69-bus system. search. This is the reason that DG locations obtained by proposed
No. of DGs Method Bus no. DG power (MW) Loss reduction (%) method are identical to ELF.
1 DG IA 61 1.90 62.91 In IA method, power factor is fixed based on load profile.
MINLP 61 1.87 62.94 Whereas, in proposed formulation, optimal power factor for each
PSO 61 1.81 64.09 DG is obtained by optimizing real and reactive power
2 DG IA 61 1.70 67.94 simultaneously. DG real and reactive powers are considered as
17 0.51 two design variables and updated iteratively by calculating the
MINLP 61 1.78 68.07 step vector dk. Optimal reactive power generation by each DG at
17 0.53
PSO 61 1.81 69.39
optimal location results in the improved voltage profile as shown
17 0.51 in Figs. 2 and 3. In 33-bus system, bus number 31 requires maxi-
3 DG IA 61 1.70 68.82
mum reactive power. The proposed formulation met this demand
17 0.51 optimally by placing DG of higher size with optimal power factor.
11 0.34 As a result, the proposed formulation achieves minimum losses
MINLP 61 1.72 69.07 with improved voltage profile. It can be observed that the proposed
17 0.38
method gives most flat voltage profiles as shown in Figs. 2 and 3.
11 0.53
PSO 61 1.81 70.08 The effect of DG units on the voltage profile for all the
17 0.51 methods with unity as well as non-unity power factor is shown
50 0.72 in Table 5.

Placement of DG units in IEEE 69 bus system

Similar trends are observed in the results given in Table 1 for mul-
tiple DG units with real power capabilities. The total real and reactive power demands in IEEE 69-bus
It is evident from the results that the MINLP based proposed system are 3.8 MW and 2.69 MVAr respectively. System line loss
algorithm shows better performance as compared to other meth- without DG is 225.27 kW. The simulation results of placement of
ods. The superior performance of the MINLP formulation is attrib- single and multiple DG units with real power are presented in
uted to the simultaneous placement of multiple DG units. In Table 6. The proposed technique gives improved performance than
addition, selective list of candidate buses by SPM model decrease IA method. However, loss reduction is maximum with PSO
the computational time and improves the search ability while technique.

Table 7
Placement of 1 DG unit with real and reactive power capability for IEEE 69-bus system.

No of DGs Method Bus no. DG power Total DG power Optimal power factor Loss reduction (%)
MW MVAr MVA
1 DG IA 61 1.839 1.284 2.243 0.82 89.68
MINLP 61 1.828 1.300 2.244 0.815 89.65
PSO 61 1.818 1.250 2.207 0.824 89.68

Table 8
Placement of 2 DG units with real and reactive power capability for IEEE 69-bus system.

No of DGs Method Bus no. DG power Total DG power Optimal power factor Loss reduction (%)
MW MVAr MVA
2 DG IA 17 0.540 0.377 2.854 0.82 96.69
61 1.799 1.2563
MINLP 17 0.522 0.359 2.765 0.824 96.80
61 1.735 1.238 0.814
PSO 17 0.524 0.371 2.749 0.816 96.69
61 1.743 1.184 0.827

Table 9
Placement of 3 DG units with real and reactive power capability for IEEE 69-bus system.

No of DGs Method Bus no. DG power Total DG power Optimal power factor Loss reduction (%)
MW MVAr MVA
3 DG IA 61 0.900 0.557 3.524 0.82 97.74
17 0.630 0.390
50 0.900 0.557
MINLP 11 0.494 0.354 3.123 0.813 98.10
17 0.379 0.257 0.828
61 1.674 1.195 0.814
PSO 18 0.5078 0.344 3.545 0.828 97.74
50 0.6996 0.474 0.828
61 1.7351 1.158 0.832
616 S. Kaur et al. / Electrical Power and Energy Systems 63 (2014) 609–617

Table 10
Critical voltages on IEEE 69-bus system with DG.

Case Method Minimum voltage pu (Bus no) Maximum voltage pu (Bus no)
UPF NUPF UPF NUPF
Without DG unit 0.9092(65) 1.0(1)
1 DG IA 0.9692(27) 0.9732(27) 1.0000(1) 1.0000(1)
MINLP 0.9682(27) 0.9724(27) 1.0000(1) 1.0000(1)
PSO 0.9681(27) 0.9724(27) 1.0000(1) 1.0000(1)
2 DG IA 0.9765(65) 0.9944(50) 1.0000(1) 1.0024(61)
MINLP 0.9789(65) 0.9943(69) 1.0000(1) 1.0000(1)
PSO 0.9806(65) 0.9943(50) 1.0000(1) 1.0020(61)
3 DG IA 0.9785(65) 0.9939(69) 1.0000(1) 1.0000(1)
MINLP 0.9790(65) 0.9943(50) 1.0000(1) 1.0000(1)
PSO 0.9806(65) 0.9940(69) 1.0000(1) 1.0000(1)

Tables 7–9 presents the simulation results for placement of Acknowledgements

single and multiple DG units with real and reactive power capa-
bility for IEEE 69-bus system. To place a single DG unit, all the The authors would like to acknowledge AICTE’s Quality
methods lead same location (see Table 7). However, the proposed Improvement Program (QIP) and the Board of Governor’s PEC Uni-
method gives maximum losses. The proposed model also gives versity of Technology, Chandigarh (India), to pursue her Ph.D.
maximum loss reduction for placement of two and three DG units work.
(see Tables 8 and 9). In IEEE 69 bus system, load demands at bus
number 11 and 61 are high. In the case of 3 DG units (Table 9), the
proposed method, due its capability of variable power factors, References
provides adequate real and reactive at those buses. The optimum [1] Keane A, Ochoa LF, Borges C, Ault G, Alarcon A, Currie R, et al. State of the art
power factor at the buses 11 and 61 are 0.813 and 0.814 techniques and challenges ahead for DG planning and optimization. IEEE Trans
respectively. Power Syst 2013;28(2):1493–502.
[2] Georgilakis PS, Hatziargyriou ND. Optimal distributed generation placement in
From Tables 8 and 9, it can be noted that the proposed MINLP
power distribution network: models, methods and future. IEEE Trans Power
based method gives superior performance for placement of multi- Syst 2013;28(3):3420–8.
ple DG with real and reactive power capability as compared to the [3] Willis HL. Analytical methods and rules of thumb for modeling DG-distribution
interaction. In: Proceeding of IEEE power engineering society summer
other two methods. The maximum and minimum voltages for dif-
meeting, vol. 3; 2000. p. 1643–44.
ferent methods are shown in Table 10. It is observed that the sys- [4] Wang C, Nehrir MH. Analytical approaches for optimal placement of
tem voltage improves in addition to loss reduction. The voltage distributed generation sources in power systems. IEEE Trans Power Syst
profile is more flat with PSO and proposed method as observed 2004;19(4):2068–76.
[5] Abu-Mouti FS, El-Hawary ME. Heuristic curve-fitted technique for distributed
from Table 10. generation optimisation in radial distribution feeder systems. Proc IET Gener
Transm Distrib 2011;5(2):172–80.
[6] Acharya N, Mahat P, Mithulananthan N. An analytical approach for DG
allocation in primary distribution network. Int J Electr Power Energy Syst
Conclusion 2006;28(10):669–78.
[7] Gözel T, Hakan M. An analytical method for the siting and sizing of distributed
This paper has presented a MINLP based approach for optimal generators. Electr Power Syst Res 2009;79(6):912–8.
[8] Hung DQ, Mithulananthan N, Bansal RC. Analytical expressions for DG
placement of single and multiple DG units to minimize the losses allocation in primary distribution networks. IEEE Trans Energy Convers
in the distribution network. To reduce the search space and com- 2010;25(3):814–20.
putational time, two-step scheme is proposed. Firstly, in Siting [9] Hung DQ, Mithulananthan N. Multiple distributed generator placement in
primary distribution networks for loss reduction. IEEE Trans Ind Electron
Planning Model (SPM), prospective candidate buses are shortlisted
2013;60(4):1700–7.
based on Combined Loss Sensitivity (CLS). This short list is then [10] Soo-Hyoung L, Jung-Wook P. Selection of optimal location and size of multiple
passed to Capacity Planning Model (CPM). In CPM, the optimal distributed generations by using Kalman Filter algorithm. IEEE Trans Power
locations and DG sizes are computed using MINLP based formula- Syst 2009;24(3):1393–400.
[11] El-Khattam W, Hegazy YG, Salama MMA. An integrated distributed generation
tion. In this formulation, Sequential Quadratic Programming (SQP) optimization model for distribution system planning. IEEE Trans Power Syst
and Branch and Bound (BAB) algorithms are integrated to handle 2005;20(2):1158–65.
discrete and continuous variables for solving the proposed formu- [12] Keane A, Malley MO. Optimal allocation of embedded generation on
distribution network. IEEE Trans Power Syst 2005;20(3):1640–6.
lation. This approach gives improved computational performance [13] Zhang X, Sharma R, Yanyi H. Optimal energy management of a rural microgrid
and strong convergence property. Due to reduced search space, system using multi-objective optimization. In: Proceeding of IEEE PES
by means of SPM model, solution converges in very less time. innovative smart grid technologies (ISGT); 2012. p. 1–8.
[14] Rueda-Medina AC, Franco JF, Rider MJ, Padilha-Feltrin A, Romero R. A mixed
The proposed methodology is implemented on IEEE 33-bus and integer linear programming approach for optimal type, size and allocation of
IEEE 69-bus test systems. A comparative analysis is done among distributed generation in radial distribution system. Electr Power Syst Res
the three popular classes of optimization methods for DG place- 2013;97:133–43.
[15] Atwa YM, El-Saadany EF, Salama MMA, Seethapathi R. Optimal renewable
ment. Comparative study in terms of DG size, distribution loss, resources mix for distribution system energy loss minimization. IEEE Trans
and computational efforts is carried out with ELF, IA, and PSO tech- Power Syst 2010;25(1):360–70.
niques. It is observed that the proposed algorithm based on MINLP [16] Ochoa LF, Harrison GP. Minimizing energy losses: optimal accommodation and
smart operation of renewable distributed generation. IEEE Trans Power Syst
gives improved performance due to its property of simultaneous
2011;26(1):198–205.
placement of multiple DG units. In addition, due to flexibility in [17] Khalesi N, Razaei N, Haghifam MR. DG allocation with application of dynamic
power factor, the algorithm gives further improved results in the programming for loss reduction and reliability improvement. Int J Electr
case of DG units capable of delivering real and reactive power. Pro- Power Energy Syst 2011;33(2):288–95.
[18] Zhang X, Karady GG, Ariaratnam ST. Optimal allocation of CHP-based
posed formulation is generalized and can be implemented for any distributed generation on urban energy distribution networks. IEEE Trans
type and any number of DG units. Sustain Energy 2014;5(1):246–53.
 S. Kaur et al. / Electrical Power and Energy Systems 63 (2014) 609–617 617

[19] Borges CLT, Falcao DM. Optimal distributed generation for reliability, losses, [27] Jain N, Singh SN, Srivastava SC. A generalized approach for DG planning and
and voltage improvement. Int J Electr Power Energy Syst 2006;28(6):413–20. viability analysis under market scenario. IEEE Trans Ind Electron
[20] Harrison GP, Piccolo A, Siano P, Wallace AR. Hybrid GA and OPF evaluation of 2013;60(11):5075–85.
network capacity for distributed generation connections. Electr Power Syst Res [28] Kansal S, Kumar V, Tyagi B. Optimal placement of different type of DG sources
2008;78(3):392–8. in distribution networks. Int J Electr Power Energy Syst 2013;53:752–60.
[21] Akorede MF, Hizam H, Aris I, Kadir MZA. Effective method for optimal [29] Wang L, Singh C. Reliability-constrained optimum placement of reclosers and
allocation of distributed generation units in meshed electric power systems. distributed generators in distribution networks using an ant colony system
Proc IET Gener Transm Distrib 2011;5(2):276–87. algorithm. IEEE Trans Syst, Man, Cybern Part C, Appl Rev 2008;38(6):757–64.
[22] Shukla TN, Singh SP, Shrinivasarao V, Naik KB. Optimal sizing of distributed [30] Abu-Mouti FS, El-Hawary ME. Optimal distributed generation allocation and
generation placed on radial distribution systems. Electr Power Compon Syst sizing in distribution systems via artificial bee colony algorithm. IEEE Trans
2010;38(3):260–74. Power Deliv 2011;26(4):2090–101.
[23] Ishak R, Mohamed A, Abdalla AN, Wanik MZC. Optimal placement and sizing of [31] Khatod DK, Pant V, Sharma J. Evolutionary programming based optimal
distributed generators based on a novel MPSI index. Int J Electr Power Energy placement of renewable distributed generators. IEEE Trans Power Syst
Syst 2014;60:389–98. 2013;28(2):683–95.
[24] Moradi MH, Abedini M. A combination of genetic algorithm and particle [32] Khatod DK, Pant V, Sharma J. A novel approach for sensitivity calculations in
swarm optimization for optimal DG location and sizing in distribution the radial distribution system. IEEE Trans Power Deliv 2010;21(4):2048–57.
systems. Int J Electr Power Energy Syst 2012;34(1):66–74. [33] Murthy VVSN, Kumar A. Comparison of optimal DG allocation methods in
[25] Celli G, Ghiani E, Mocci S, Pilo F. A multiobjective evolutionary algorithm for radial distribution systems based on sensitivity approaches. Int J Electr Power
the sizing and siting of distributed generation. IEEE Trans Power Syst Energy Syst 2013;53:450–67.
2005;20(2):750–7. [34] Leyffer S. Integrating SQP and BAB method for mixed integer programming.
[26] Gomez-Gonzalez M, López A, Jurado F. Optimization of distributed generation Comput Optim Appl 2001;18:295–309.
systems using a new discrete PSO and OPF. Electr Power Syst Res [35] Fourer R, Gay DM, Kernighan BW. AMPL: A modeling language for
2012;84(1):174–80. mathematical programming. 2nd ed. Brooks/Colo-Thomson Learning; 2003.