National Power 6r Encrgy Confcrcncc (PECon) 2004 I’rocccdiiiEs,
2KX
Optimal Allocation and Sizing of Embedded
Generators
T.K.A. Rahman, S.R.A. Rahim and I. Musirin
Absfrucf - Recent years, the nuniber of enibcdded generators
installed in distribution system has been increasing in many
parts of the world. Depending on their operating
characteristics and locations, embedded generators could
significantly affect the voltage profile, network losses and fault
level in a distribution system. This paper presents a new
tcchniquc for dctermining optimal allocation and sizing of
enibcdded generator in a distribntion system. Sensitivity
indices based on voltage stability improvement with rcspect to
change in injected active and reactive powcr at a load bus were
derived and used to identify the suitable location for the
embedded generators. In order to determine the optimal
output of the embcddcd generators, an evolutionary
prograniining optimization technique was developed with 311
objective to niininiizc thc distribution losses wliilc satisfying the
voltage constmint in the system. The proposed tecliniquc was
tested on ttic 69 bus distr4butioti systcm and the results shown
a significant redtietion in distribution losses and voltage profile
improvement in the system with the implementation of thc
cinbcdded generation nt the snitnble location m t l optiinsl
sizing.
Itrda Ten)i - Enibedded Generation, loss minimization,
sensitivity analysis, Evalutioiiary Programming.
I. lNTROUUCTION
Rapid industrialization process and population growth have
resulted in an escalation in tlic clectrical powcr
consumption. The limited area and slow progress in network
expansion lias also caused areas with high load densities.
These may introduce power quality and voltage stability
issues[ 11. At the same time, rural electrification networks
are cxpericncing poor network performance in ternis of
large voltagc drop and high distribution losses along the
lines. Distribution utilitics arc trying very hard to strengthcn
and expand their networks with limited source from the grid
and also capitai. Introducing embedded generation to the
distribution network could be the answer to these problems.
In the recent years, embedded generation Iias shown an
increasing growth in distribution networks around the world
due to rise in promotion towards utilization of renewable
energy resources and development of co-generation plants.
With proper planning, the integration of embedded
geiierations in a distribution system would lead lo
T.K.A. Rnhmm, S.R.A. Raliim and I. Musirin are with Faculty of Electrical
Engineering, Univcrsili Tcknologi MAKA, 40450, Shall hlam, Sclangor
Darul Ehsaii, MALA YSlA (e~iicril:takilikii7!tiii.nct.my)
0-7803-8724-4/04/$20.00 02004 IEEE.
Ku;ilti Lumpur, Mnloysia
enhancement in the network performance in ternis of
voltage profile improvement, reduction in line losses and
improved power quality [2,3]. As a result, tlie deinand
required from the grid could be reduced, thus cutting the
need to strengthcn tlie feeders connecting thc network to tlie
grid.
Many studies have been performed to identify tlic influencc
of embedded generation on the perforniance of a distribution
network. The effect of embedded generation on thc voltagc
profile in a distribution network was examined with rcspect
to both distributcd and luiiipcd loads [3]. Thc study rcvealcd
that the uppcr limit valuc of the power that can bc injected
into a line without causing ovcr voltages are: tlic voltagc at
the beginning of tlic line, the upper voltage limit, thc
position of tiic iiijcctcd point, the total current drawn by thc
fccdcr loads and h e quantitics rclatcd to thc load currcnt
distribution. Study has shown that tlic inclusion of
synchronous or induction embcddcd gcncrators 1iavc
modified thc behavior of a distribution nctwork during
transicnt and stcady statcs[l]. Thc bcst advantage in tcmi of
voltagc stability was gained through collcctivc integration of
both typcs of embcddcd gcncrators. licrercncc 4
demonstratcs tlic application of load control for voltagc
rcgulation i n distribution fccdcrs with cnibcdded wind
generation. This sclicnic is more atiractive than thc
conventional volhgc control tcchniqucs sincc it is capablc to
perform pcak looping, allcviating undcr voltagc problcm
and avoiding liiic current ovcrloads. I n addition, rcfercnce 5
shows that cmbcddcd gcncrator opcrating on voltagc control
mode has able to control thc voltage profiic of the nctwork.
Bcsidcs improving tlic voltagc along the fecder, a significant
reduction in tlic linc losscs could be obtaincd when an
embcdricd gcncrator is properly sized to closcly match and
located to tlie local load [GI. Thcrefore dctermining thc
optimal sizing of the enibcdded generator is onc of the
aspects presented in this paper.
Many coiiventional optimization techniques such as the
gradient niethods, lincar programming, quadratic
programming and dynamic programming have been
employed to solve power system optimization problems in
system planning, opcratjon and pricing. However, duc to the
complexity of the problems, these methods may fail to find
the global optimal solutions [7].
Evolutionary computing techniques are cmcrging as
efficient approaches for various search, classification and
 289
optimization problcins. Tlicsc C ~ ~ S S C Sof computing
algorithms are callcd Evolutionary Algorithms ( E h ) which The first sensitivity index was formulatcd from the change
arc the cotnputcr-bascd problcin-Solving systcnis basctl on in L, with rcspcct to the change in injected P, at bus i is
the principlcs of cvolutionary theory, Thc most popular EAs given by
developcd are :-
a) Genctic Algorithms(GAs)
b) Evolutionary Programming(E1')
c) Evolutionary Strategies(ES)

The interest in these algorithms has been rising fast. The Hence, the second sensiiivity index was derived from the
main strength of these innovative search algorithms lies in change in L with respect to the change in injectcd Q at bus i
their global convergence, inherent parallel processing is given by
nature, problem independence and great robustness.
Evolutioiiary computing has shown to be providing an
impressive solution to VAR planning problems in
transmission and distribution networks. Genetic Algorithms
has been implemented to determine the size, location, type
and number of capacitors to be placed in a radial distribution Expressing equations 2 and 3 into matrix form gives
network in order to minimize the network losses [8,9].

In this paper, a new method for determining the suitable

location for embedded generator is proposed bascd 011 new ......(4)
sensitivity indices derived from voltage stability
improvement with respect to changes in injccted active and
rcactive powcr at a bus. Evolutionary programming
optimization technique has been developed in order to
dcterniine the optimal sizc of the enibcdded generation. Thc
objective of the optimization was to minimizc thc losscs in
the network wliiie maintaining thc voltagc profilc at the
acceptable level. The effcctiveness of the proposed
methodology was verified by thc aiialysis 011 a 69-bus
distribution system[ IO].

Tlic clcments of the row matrices in equations 4 and 5 are

11. SENSITIVITY INDICES derived from equation L as follows,

In order to obtain maximum benefit froni the einbeddcd

generator, suitable location and sizing lias to bc dctcrmined
before its installation. The proposed tcchnique identifies thc
suitable location for the embeddcd generator by studying the
prc-developed voltagc stability index [ I I ] variations with
respect'to changes in reactive and active powcr injections at
a bus. Sensitivity indiccs that relatc the changes in thc
voltagc stability index with rcspcct to changes in injected
active and reactive power at a load bus were derived from
inverse of loadflow jacobian matrix.
the voltage stability index formulation. The sensitivity
indices were computed for every load bus and those buses
with highest sensitivity values were choscn for the The sensitivity criterion was determined from the vahes of
embedded generation placement. The derivation of the the sensitivity indices evaluated at each load bus in a
sensitivity indices from the voltage stability index system. Buses with highest sensitivity values are selected for
formulated in reference 1 1 is as follows:- the location of the embedded generators.

The voltage stability index at a load bus i i s given by

Li = 4[YOlVLi cos Qi - Y t cos Q i 2 ] I
VL, = load voltage at bus i
V,, =no load voltage at bus i
Qi (60, - OL,1
O,, = load angle at bus i
O,,= no load angle at bus i
111. EVOLUTIONARY PROGRAMMING (E")
v; ................(1) Evolutionary Programming has been employed in the field
of design search and optimization more thoroughly after the
exposure from Fogel [12] when it was first implemented in
the prediction of finite states machines. Since then, EP has
undergone refinement process in which self-adaptation
parameters and different mutation strategy has been
implcmented. EP searches for the optimal solution by
evolving a population of candidate solutions over a numbcr
of gcncrations. During each generation, a new,popiilation is

formed from !he existing population by implementing the
mutation operator. The operator produccs a new solution by
perturbing each component of the current solution by a
random amount. The strength of cach o r the candidate
solution is determined by its fitness that is evaluated from
the objective function of the optimization problcni.
The selection process is done through the tournament
scheme, in which individuals from a population compete
with each other. The individuals that obtained the most
numbers of wins will be selected for the new generation.
The competition scheme must be such that the fittest
individuals will have a greatcr chance to survive, while
weaker individuals will be eliminated. Through this, the
population evolves towards the global optimal solution.
In the proposed approach, Meta EP [I21 technique was
employed for determining the optimal size of the embedded
generator. This technique is different from the standard EP
since element of self adaptation is included in thc mutation
process. The implementation of Meta EP was according to
- .proccdurcs:-
the followinF:
a. Generate the initial populations of a consisting of i t
individuals and set generation K = I . Each
individual will be from a pair of real vaiued vcctor
(xi , q;), V i E { 1,..,, p}, whcrc q i is a strategy
parametcr.
b. Each a has I I components a()),j I,.. . , ) I
c. Evaluate the fitness score for each individual (x,,
qi), Vi E { l , ....
p}, of the population according to
the objective fhctioti,f(xJ
d. For each parent ( x i , vi), i = { l ,.... p }, a singlc
offspring is creatcd (xi’,vi‘) using:
TJ,~(~} = q 0;)
rN,(O,I)), .............@)
x, 0;) = xi 0;) i-
..................... (9)
xi tj),xi ‘G), vilij and is the jlh component
of the respective vector. N(0,i) donates a
normally distributed one-dimensional random
number with mean 0 and 1. N,{O,I) indicates
that the random number will be anew for each
value ofj.
r = ( (2 (n)”2)”’)-’ and
.............(10)
e. Calculate the fitness of each offspring (xi,qi’) i =
11 .>PI.
I . .
f. Pair wise comparison is conducted over the union
o f both parent and offspring. For each individual, q
opponents are chosen randomly from the union
with equal probability. In cach comparison, an
individual receives a “win” if its fitness number is
not smaller than the opponents.
290
The p individuals out of the union that liiivc thc
niost wins will bc selected as parcnts of ncxt
generation,
Stop if the halting creation is satisficd, otherwise
go the next generation, k = k -F 1 and continue at
step c.
The implementation of EP for determining the optimal size
of embedded generator in order to minimize the line losses
in a distribution system is explained in the next section.
IV. OPTIMAL EMBEDDED GENERATOR
SIZING VIA EVOLUTIONARY
PROGRAMMING
The optimal size of the embedded generator is determined
by having the kW output (Pg) of the embedded gencrator as
the variable to bc optimized in the EP optimisation. The
kVar output of the embcddcd gcncrator was determincd
using equation 12 and the power factor o f the system is set
to be 0.85.
Y, ‘ P g ........................................ (1 1)
es = P~’PX. tan-‘ 4............................ (1 2)
= 0.85
COS#
4 = power factor anglc
The operation of the embeddcd gcncrator is considcrcd 10 be
at steady state and therefore, the cmbcddcd generator is
modeled as injected active and rcactivc powcr, P, and Q,
exp(iNfO,Ij f respectively [ 131.
The objective of the optimization is to minimizc the network
17; ()I qi‘O,l), losses denoted by equation 13. Hence, the fitness for the EP
was taken to be the total losses in the distribution systcm
and evaluatcd by executing the loadflow programme with
the injected active and reactive powcr at thc suitable
location determined from the sensitivity analysis. The
optimization also took into consideration the voltage
constraint of the system as shown in equation 14, so as to
ensure that the inaxiiiiuni and minimum voltagcs would not
be exceeded.
7 ‘= ( (2 TI)''^)-' )I
Miininiisc~P,,, ................................... 13)
11
’
n = numberof linesin thesystem
Voltageconstrain&, ...(14)
VI,,;,, < V, < V,,,,, ......_.._
The population size is taken to be 20 and the mutation
probability is 0.03. The procedure of thc EP nicthod to
detcrniitie thc optimal size of the embcdded gcncrator is
sunimarized i n the flowchart given in figure I .
 29 1
v. SIMULATION RESULTS AND DISCUSSION

This study investigated on the loss minimization and voltage

improvement that could be achieved as a result ofembedded
generator installation at the chosen bus with optimal

P-r-4
generation output identified by the EP optimization Generate initial population (x,, q )
technique. The developed technique was tested on a 69-bus
system[ IO]. This system contains 69 branches and 7 laterals
as shown in figure 2.

In order to determine the suitable location of the embedded

e? Satisfy Constraint?

generator, sensitivity analysis was conducted on the 69-bus

test system. The sensitivity indices given by equations 4 and
5 were evaluated for every bus in the system and the results
are tabulated in Table 1. The bus with highest sensitivity
index value is selected for allocating the embedded
[A,
i"rt in population pool

generator.

From Table I , it could be observed that bus 61 has the

highest sensitivity index value and therefore it is chosen as
the suitable location for embedded generator. However, for
comparison, buses 64?21, 65 and 5 9 were also selected for
,A, Evaluate fitness vaiue

embedded generaror allocation so that the improvement on

the network performance in terms of loss minimization and
Mutate x,, q ,
voltage profile improvement could be compared. The study
also considered the effect of Ioad increment at the chosen
buses individually and also load increment at all buses in the
-7-
system simultaneously on the network losses and also
voltage profile. The EP optimization technique was
implemented in order to determine the optimal output of the
embedded generator located at the chosen site.
Add into population pool
and evaluate fitness
The graphs in Fig. 3 and Fig. 4 show the variation of total
losses and voltage profile in the system respectively with
respect to load increase at buses 61, 64, 21, 65 and 59
individually.
Clone population IS ,

full 7
From the graph shown in figure 3, it could be observed that
load increase at bus 61 has caused an increase in the total
losses of the system significantly. At the same time, figure 4
shows that increase in load at bus 61 has also reduced the Merge with the old
minimum voltage appreciably as compared to the effect of population
load increase at the orher load buses.

Selection for next

generation by pairwise
tournament
1
2 Convergence Test

I stop
1
Fig. I . Flowchart for implrmemation of EP

.T-
 292

28 29 30 31 32 33 34 35
1 1 1 1 I I I
I I I I I

I l l l l l l l l l f
1 1 1 1 1 1 1 1 1 1 1
6 17 18 19 20 21 22 23 24 25 26 27

,
1
i
1
$6167,
1
I
1 1 1 1 1 1 1 1 1
1 I I 1 1
6 57 58 5960 61 62 63 64 65

47 48 49 50

Fig. 2. The 69-bus Tcst Sysicm

TABLE I
FIRST 10 BUSES WITH HIGHEST SENSITIVITY INDEX VALUE

. _ _., .. ,.
. .
.
badinp
. .

in loading ill rcspcclivc bus rndividually.

In ordcr to niinimizc thc systcm losses, embcddcd gcnerator

was located at these rcspcctivc buses individually. Thc loading
at cach load bus was varied and the EP optimization technique
was implcmcntcd to dctcrminc the optimal embedded gcncrator
output in ordcr to minimizc thc systcni losscs at the various
loading conditions. The graph in Fig. 5, illustrates thc
distribution losses in the system with embcddcd gcncrator
installed at respcctive load buses for a range of loading
conditions. Fig. 6 shows the minimum voltage in the system as
a result of installing embcdded generator at the respcctive bus.
The optimal output of the embedded generator in order to
minimize the system losses at different loading condition
identified by the proposed EP optimization tcchnique are
tabulated in Table 2.
 193
loading as shown in Fig. 6. Therefore, the results demonstrate
that,the proposed EP optimization technique has able to search
@fofi&e;?Q$@l output of the embedded generator in order to
minimize the system losses while maintaining the.voltage level
at various loading conditions. The resuits tabulated in Table 2
also reveal that, as the loading at a load bus was incremented,
the optimal output of the embedded generator has also
increased. The power to be generated by ,the embedded
generator located at bus 61 i s higher as compared to other bus
Iocation. This is due to the high loading connected at this bus.
However, allocating the embedded generator at bus 61 has
resulted in the greatest loss minimization and voltage
improvement.

Fig. 5 . Total losses in the system with installation ofembedded penemtor

for load increase at individual buses.

overall load increase in the system.

Optimal output of Embedded generator at each bus

~~~d~~~ location (MW)
61 6 4 1 59 I 65 I 21
0.6 I 1.3428 1 1.5263 I 1.8895 I 1.3901 I 0.7312 I
I 0.8 1 1.5910 1 1.5731 1 1.9086 I 1.3996 1 0.7525 1
1 10 I 1.8393 I6172 I 1.9272 I 1.4144 I 0.7782 1
t.2 2.0878 1.6643 1.9482 1.4260 0.7970
generator for overall load incrense in the system.

1 1.6 1 2.5856 1 1.7546 1 1.9856 1 1.4496 I 0.8450 I

/ 1.8 I 2.8349 1 1.8008 2.0085 I 1.4609 I 0.8681 I
Fig. 5 shows that allocating the embedded generator at bus 6 1
has resulted in better loss minimization as compared to other
bus location. The minimum voltage in the system has also been
maintained at an acceptable value despite the increase in
 294
TABLE 3 distribution syslom. The proposcd tocatioii o f llic cmbcdclcd
OPTIMAL EMBEDDED G E N E R A T O R O U T P U T FOR LOSS gcncrator Iiiis iilso ablc 10 ~iroducch c bcst rcsult it] tcrins of
MINIMIZATION I N THE S Y S T E M FOR 0VERAL.L LOAD 1NCREASE IN
THE SYSTEM voltagc improvcnicnl at various loading condition with optimal
Optimal output of cnibcdded geiicrstor at cacti bus output idcntificd by llic dcvclopcd EP optimization technique.
~ ~ location
~ (MW)
d i ~ ~ ~
61 64 59 VI I. REFERENCES
0.6 1.1009 I 0.9632 I 1.1464
J. V. Milarrovic uird T. M . Diivid, "Stubility of Distribution nciworks with
Embcddctl Gcncraiors uiid Iiiduclion Motors". /EEE Power. Eilgineering
Socio!)~Wirilsr Meeliiig 21102, Vo1.2, pp 1023 - 1028, Jan. 2002.
D.Xti aiid A. A . Giryis, "Optiinal load shcddiiig strategy i n powcr
systcms will1 dislribuicd gciicration", E E L Poiwr ErigDiecrbig Societ)~
Wiritci, Medrig 2001, Vo1.2, pp 788 - 793, Fcb. 200 I,
S. Coiiti. S. h i l i iliid G. Tina, "Siiiall-scalc cmbcdded gancration cffcct
on voliiigc prolilc: iiii uiialylicnl method", I'roc. /.EL: Got. Trotis, Divt..
vol. 150, no, I, pp. 78 - 86,Jim 2003.
The capability of the proposed tcchnique to detcrininc thc
optimal output of the embedded generator i n order to minimize
thc total losses i n the system with overall load increase in thc
system was tested. A range of loading condition (60% - 180%
of the nominal value) was used for the evaluation and thc
results are given in Fig. 7, Fig, 8 and Table 3. Thc graph shown
in Fig, 7 compares the total losscs in the system bcforc and
after thc installation o f emhcddcd generator at the selccted bus
location. The graph also shows that the total losscs in the
system were rcduccd with the installation of emhcddcd
generator and the best loss niinimizatioii is obtained wlieii tlic
embedded gcncrator was locatcd at bus G I . Similarly, tlic grapli
in Fig. 8 shows the variation i n tlic minimum voltage with
respect to overall load increase in the system. The graph also
illustratcs effect of cinbeddcd gencrator installation at [lie
selected buses on voltage iiiiprovemcnt. It could be observed
that allocating embedded generator nt bus 61 has given bctlcr
voltage improvement and Iiencc tlic voltagc profile of the
system is maintaincd at an acccptablc range. Finally thc
optimal active power to be gcncratcd by the ctnbedded
generator installed at the sclectcd bus location for a range of
loading conditions are tabulatcd in Tahlc 3 . It could bc
observed that more injectcd powcr is requircd from the
embedded gcncrator as the loading was increased in ordcr to
minimize thc losses and also improving the vol~ageprofilc in
the system.

VI. CONCLUSIONS

A new approach for dctermining the location and sizing of

embedded generator in a distribution system is prescnted. Two
sensitivity indices were formulated based on voltage stability
improvement with respect to changes in active and reactive
power injection at load bus. TIE sensitivity indices were
cvaluated at eveiy load bus and the bus with highest sensitivity
index valuc was chosen to bc the suitable location for the
cnibedded generator. An EP optimization technique was
developed to deteriniiie the optimal ,output of the embeddcd
generator. From the n~rinericalsimulation, the combination of
suitable location of the embedded generator identificd from thc
sensitivity analysis and optimat generation output idcntified by
the EP optimization technique has ablc to rcducc the losses in a