1

Optimal Location and Capacity of Distributed

Generation Based on Scenario Probability
Ding Xiaoqun, Wu Jiahong, and Zhao Feng

Many studies have been performed to plan distributed

Abstract—The paper analyzes distributed generation generation [6] - [10].
planning of capacity and location. Distributed generation In [6], an optimal reinforcement strategy to provide reliable
usually installed in distribution system. Distributed generation and economic service to customers in a given time frame is
such as solar generation and wind generation power output is
investigated. The DG technology is considered as a solution
affected by meteorological conditions. DG power production
fluctuates frequently, DG injected distribution system is seldom for distribution systems capacity problems. Compound
dispatched and controlled by operators. Some control means solutions comprising the installation of both feeders and
could not adapt to operation mode frequently varying. substations reinforcement and DG integration at different
Distributed generation planning method must adapt to DG times are considered in the formulation of a minimum cost
power output random large-scale change. Scenario Probability distribution systems reinforcement strategy problem. The
methodology is applied to take DG optimal planning. The
proposed technique is the Dynamic Ant Colony Search
methodology separates random DG output into several portions.
Every portion is related to different DG output scenario. Every algorithm. In [7-8], from the perspective of a distribution
scenario happening probability is estimated. Planning methods company, propose a new heuristic approach for distributed
also consider investment cost and power loss of distribution generation capacity investment planning. The model aims to
network. Technical constraints such as feeder capacity limits, minimize the distribution company's investment and
feeder voltage profile are considered. PSO algorithm is applied operating costs as well as payment toward loss compensation.
to solve planning problem. Example of IEEE-33 shows that the
In [9], DG offers an attractive alternative to T&D expansion.
proposed method is feasible. Planning DG embedded in
distribution system can well adapt DG output fluctuation. Instead of expanding existing substations, building new
transmission lines, and building new substations, DG can be
Index Terms-- Distributed Generation, Distribution network, used to accommodate new load growth and provide relief for
Scenario probability, PSO algorithm, Planning model. overloaded components. In [10], Sensitivity indices based on
voltage stability improvement with respect to change in
I. INTRODUCTION injected active and reactive power at a load bus were used to

D ISTRIBUTED Generation(DG) is a small generator identify DG suitable location.

scatter throughout a power system. Due to locally Distributed generation has many sorts. Some new energy
available resources and the small scale, DG units are mostly generation such as wind turbines, photovoltaic and other
installed in demand system and directly connected to clean natural energy can reduce environment pollution. New
distribution system [1]-[2]. Distribution generation has many energy DG has no demand to fossil fuels compared to
types including wind turbines, photovoltaic, fuel-cells, conventional power plants. DG set up in the vicinity of the
biomass, micro turbines, small hydroelectric plant, etc [3] - customer also reduces power transmission losses.
[4]. Conventional power plants power output are stable. However,
As deregulation in power systems, environmental impact new energy DG output power fluctuates frequently on
of electric power generation, several generation technologies account of meteorological conditions alterability [11]. For
advancement as well as rapid increase in electric power example, wind turbines output power is related to wind speed,
demand, numerous DG were connected to distribution wind speed strongly wave give rise to wind turbines power
networks. Distribution generation systems can reduce the output large-scale change randomly. In view of these sorts of
system loss and defer investment on transmission and natural energy DG output power swing obviously, and DG
distribution expansion. Appropriate size and optimal injected distribution system is seldom dispatched and
locations are the keys to achieve it [5]. controlled by operators, some control means could not adapt
to operation mode of steep fluctuation. So DG connected into
distribution system planning method should consider natural
Ding Xiaoqun is with the Department of Electrical Engineering, Hohai
University (e-mail:ai_group@163.com). energy generation randomly fluctuating output. Planning
Wu Jiahong is with the Department of Electrical Engineering, Hohai scheme should be suitable to various operational modes.
University (e-mail:hhuwjh@163.com). Technical constraints such as feeder capacity limits, feeder
Zhao Feng is with Jiyuan power supply company, Henan electric power
company.
voltage profile are required. Traditional power planning
 2

methods according to rated capacity couldn’t meet various values associated with scenario probability. The approach
operational modes. DG planning results likely adapt to some may be described as (1):
operation mode, other operation modes are probably worse
off inversely. The key of planning natural energy DG sitting
{ }
m i n or max E ⎡⎣ F ( Si ) ⎤⎦ =

and sizing is how to adapt DG power output randomly. ⎛ n

⎞
min or max ⎜ ∑ Pk Fik ⎟ ⇒ Sopt (1)
Considering uncertainties on DG power production, this ⎝ k =1 ⎠
paper used the methodology of probability of scene
occurrence [12]-[13]. Different scenarios of power where n is alternative scenario number, pk is probability of
production have to be considered. Each scene is associated to
the k th scenario, Fik is objective value in the i th alternative
a different DG power output. Every scenario happening
probability is estimated. Planning methods also consider schemes when the k th scenario happened. E represent
investment cost and power loss of distribution network. mathematical expected.
Technical constraints such as feeder capacity limits, feeder Equation (1) represents the best objective on the average.
voltage profile are in consideration. Particle swarm
optimization (PSO) algorithm is applied to solve planning III. PLANNING MODEL FOR DG
problem. Example of IEEE-33 distribution system shows that DG planning model involve DG equipment investment
the proposed method is feasible. Planning embedded DG costs, power loss, voltage quality and technical constraints
distribution system can adapt DG output fluctuation. such as feeder capacity limits, injected DG total upper
capacity.
II. OBJECTIVE ASSESSMENT METHOD BASED ON SCENARIO A. Investment costs
PROBABILITY
The costs of DG investment consist of equipment purchase
Scenario probability methodology is introduced to resolve costs and installation costs. In order to consider time effect of
uncertainty problems in planning. Scenario decision theory money, total investment fees will be equivalent to investment
reflects all uncertain aspects related to planning result. Every cost per year, as in (2):
part corresponds to a certain scenario. Then certainty
⎛ nnode ⎞ r (1 + r ) m
measures are taken to deal with every scenario. Finally, an min ⎜ ∑ ci xi ⎟ m ( 2)
estimate method is given to meet planning demand. ⎝ i =1 ⎠ (1 + r ) − 1
Essentially, this methodology translates uncertain factor
which is hardly expressed in mathematic models into multi where: r is discount rate, m is equipment service years, ci is total
certain scenario problem to resolve. It minimized
complicated degrees of uncertain problems. investment cost per DG unit, xi is installed DG number in the
The key of scenario probability methodology is to i th bus , nnode is the number of bus.
reasonably select related typical scenarios. Uncertain aspects
which affect planning results have many factors including B. Network loss
power output, load, equipment fault, maintenance and Network loss is also an important aspect in planning
operation costs, et al. In order to highlight the influence of objective. It is generally assumed that electric power always
DG power fluctuation to DG planning, this paper has been flows from the secondary winding of the transformers in the
hypothesized that the power production of DG is the only substations to the end of feeders. Whereas, DG integration to
uncertainty elements. The Uncertainties in power production network brings bidirectional power flows. In general, DG
of DG units are represented by a set of scenes. Each scene is offers loss reduction benefit. However, this benefit cannot be
associated to different value of DG output. Natural energy taken for granted [15]-[17]. Network loss can be gained by
DG power output is determined by meteorological conditions. A power flow calculation.
typical case is that the power of wind turbines depends on the wind
speed random behavior in the area where wind turbines are C. Constraints
installed. Therefore, Meteorological distribution of planning sites The equality constraint is power flow equation. The
needs to be investigated. Although meteorological conditions are in inequality constraints include voltage upper and lower limit,
uncertainty, meteorological data generally takes on statistical
characteristic [14]. So the probability researches of DG output the maximum feeder transmission capacity limits, and the
scenario have to concern about meteorological change daily and maximum level of DG penetration rate.
monthly. It should be noted that the meteorological statistic not only voltage constraints:
should pay attention to collect history data, but also carry out actual V min ≤ Vi ≤ V m ax ( 3)
area meteorological data measurement.
After identify typical scenario and scenario probability, feeder transmission capacity constraints:
another important issue with respect to application of Pj ≤ Pj max ( 4)
scenario decision is how to determine the best comprehensive DG penetration capacity constraints:
scheme. That is, how to choose planning objective. General nDG
objective function is to take minimum mathematical expected ∑P
i
DGi ≤ η Pload ( 5)
 3

where V is node voltage, Pj is j th branch transmission xik +1 = xik + vik +1 (9)

capacity, Pload is total load level, η is DG maximum k
i
k
where v and x are velocity and position of the i particle in
i
th

penetration rate, nDG is total number of DG. k th iteration, respectively; c1 and c2 is study factor, each value
D. Single Scenario Objective function is set to 2; ω k is weight function, it is used as following:
The general formulation of a single scenario planning model ω − ωmin
ω k +1 = ωmax − max ×k (10)
is expressed: iter
⎛ nnode ⎞ r (1 + r ) m where ωmax and ωmin are the maximum and minimum weight
f = min { ⎜ ∑ ci xi ⎟ m
+ ω PLoss +
⎝ i =1 ⎠ (1 + r ) − 1 respectively, ωmax = 0.9 , ωmin = 0.4 in this paper. k and
nnode
iter are the k th count and maximum iteration count,
μ1 ∑ [max( vi − vmax , 0) + max( vmin − vi , 0)] +
i =1 respectively.
nbranch In this paper, each node’s DG injected capacity is
μ2 ∑ max( P − P
i =1
i i max , 0) } ( 6) represented by particle:
⎡ x11 x12 x13 ... x1N ⎤
where μ1 and μ 2 is punished factor, ω is weight coefficient of ⎢x x22 x23 ... x2 N ⎥⎥
power losses. P = ⎢ 21 (11)
⎢ # # # ... x3 N ⎥
E. Multi-Scenario Objective function ⎢ ⎥
⎣ xM 1 xM 2 xM 3 ... xMN ⎦
The synthetic indexes of planning model involved a set of
where P is particle population, M is population size, N is
possible scenarios:
total number of nodes, xij is the j th node’s total DG capacity
⎛ n ⎞
{ }
m i n E ⎡⎣ F ( Si ) ⎤⎦ = min ⎜ ∑ Pk f k ⎟
⎝ k =1 ⎠
(7) in the i th particle.
The procedure of planning by PSO is shown in Fig.1:
where n is alternative scenario number, pk is probability of
the k th scenario, f k is objective value of the k th scenario Input distribution system data ；
according to (6). PSO algorithm parameter

IV. PARTICLE SWARM OPTIMIZATION ALGORITHM Randomly initial velocity and position
of all particles ; Set iteration counter
FOR DG PLANNING K=1
Many conventional optimization techniques such as the
gradient methods, linear programming, quadratic Print out location and
Y

K>Kmax ?
programming and dynamic programming have been size of DG ， End

employed to solve power system optimization problems in N

system planning. However, due to the complexity of the Run the power flow in selected every
problem, these methods may fail to find the global optimal scenario; obtain voltage , loss and
other data
solution [18].
Particle swarm optimization (PSO) is an efficient approach
Evaluate objective function (7)，
for optimization problems. It is proposed by Kennedy and Record Pbest and Gbest
Eberhart in1995, inspired by social behavior of bird flocking
[19]. The PSO as an optimization tool provides a population-
based search procedure in which individuals called particles Update particle position and velocity
change their position with time. In a PSO systems, particles
fly around in a multidimensional search space. During flight,
each particle adjusts its position according to its own Y

Stopping criteria ok ?
experience (this value is called Pbest), and according to the
experience of a neighboring particle (this value is called N

Gbest), made use of the best position encountered by itself

K=K+1
and its neighbor.
Velocity of each particle can change by the following Fig. 1 Flow chart of DG planning by PSO
equation:
vik +1 = ω k v ik + c1 × rand (1) × ( Pbest ik − xik ) + V. SIMULATION
In this section, IEEE-33 bus distribution systems are
c2 × rand (1) × (Gbest gk − x ik )) (8 ) applied to test the methods proposed in above paragraphs.
Particle position update can be expressed as: Test systems specification is narrated in [20], systems
 4

structure is shown in Fig.2. The DG is planned into this Optimal results are shown in Table I. Results include
system. System voltage is 12.66kV. Substation bus 0 is slack installed DG’s location and the number of equipment in
node. The original total real power and reactive power loads relevant site. The number in bracket is the number of
are 3715kW and 2300kvar. In planning level year, the growth installed DG corresponding to the site of load bus.
rate of power demand is 5%. The total planning capacity of
DG is no more than 25% of total loads. Voltage normal range TABLE I
RESULTS OF DG LOCATION AND SIZE
is permitted from 0.95p.u to 1.05p.u. Type of DG is chose as
constant power factor (0.95) wind turbine. Rated power of
Scheme Results
each DG unit is 100kW. Investment cost is 20 ×104$ per wind
1 16(1), 17(1), 31(3)
turbine. DG serving time is 20 years, discount rate is 0.1. The
2 15(1), 16(1), 17(1), 31(3)
punished factor μ1 and μ 2 is set to 1000，weight coefficient
3 15(1), 16(1), 17(2), 30(1), 31(4)
are determined as ω =12 by trial and error ，in equation (6). 4 17(3), 30(1), 31(4)
18 19 20 21
25 26 27 28 29 30 31 32
Scheme 1, 2 and 3 used single scenario of power output to
optimize DG’s capacity and size. Scheme 1 took DG power
output as rated production. Scheme 2 considered that DG
power output is the 75% of rated capacity. Scheme 3
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 regarded DG power output as 50% rated capacity. Scheme 4
made use of synthetic indexes to optimize planning. Method
22 23 24
of Scheme 4 involved a set of possible scenarios effecting
Fig. 2 33-bus distribution system planning result.
From Table I, we can found that the sites of DG are all
In order to better explain the proposed methodology in close to end of radial network’s feeder. It is concluded that
section II, the case of 3 scenarios has been considered: DG connected to end of feeder can better enhance voltage
Scenario I: DG output is equal to rated power; Scenario II: level.
DG output is 75% of rated power; Scenario III: DG output is It is important to notice that 3 approaches lead to different
50% of rated power. The probability of each scenario is 0.48, choice results. Scheme 1 has 5 DG; Scheme 2 has 6 DG; 9
0.39 and 0.13, respectively. DG in scheme 3; 8 DG in scheme 4.

TABLE II
THE COMPARISONS OF FOUR SOLUTIONS

Scenario I Scenario II Scenario III

Scheme Power Loss The number Power Loss The number Power Loss The number
(MW) of abnormal (MW) of abnormal (MW) of abnorma l
voltage bus voltage bus voltage bus
1 0.1437 0 0.1591 6 0.1759 10
2 0.1324 0 0.1495 0 0.1688 9
3 0.1039 0 0.1245 0 0.1496 0
4 0.1130 0 0.1327 0 0.1560 0

The comparisons of four solutions are shown in Table II. Scheme 3 can adapt well to three scenario operation.
Table II shows, using planning results of Scheme 1 to Network loss is smallest, whereas cost of investment is in top
connect DG units to test network, network loss is 0.1437MW, rank.
no voltage is out of limits when DG units run in Scenario I, Although Scheme 4 didn’t take on optimal cost and
namely rated output state. However, when Scenario II and network loss on account of each scenario probability in
Scenario III happened, the number of abnormal voltage nodes different scenarios, this approach solution is the best on the
is six and ten, respectively. It is shown that scheme 1 average of future.
couldn’t adapt to wide-range power output fluctuation. From Scheme 4, we can notice that all constraints can be
Using Scheme 2, test systems can normally operation all satisfied, no matter which scenario happened. Scheme 4
when has smaller loss and better voltage quality compared with
DG output changes from 75% to 100% of rated power. But Scheme 1 and 2.
there still exist nine abnormal bus- voltages for Scenario III. Scheme 3 has lower loss than Scheme 4, but investment
cost rise. Moreover, Scheme 3 occurring probability is lower,
 5

so investment costs looks more economic compared with loss [13] Van Geert, Edwin, "Increased uncertainty a new challenge for power
system planners," IEE Colloquium (Digest), 1998, n 200, 1/1-1/6.
index of Scenario III.
[14] P B Barker, R W de Mello," Determining the impact of distributed
Based on above discussion, the synthetic indexes of generation on power systems: Part I-Radial distribution systems," In
planning model which considered a set of possible scenarios Proc. Of 2000 IEEE PES Summer Meeting, vol. 3, pp. 1645-1656.
is average optimal, can adapt well to large-scale DG power [15] P B Barker, R W de Mello," Determining the impact of distributed
generation on power systems: Part I-Radial distribution systems," in
fluctuation. Moreover, DG units which are injected into
Proc. Of 2000 IEEE PES Summer Meeting, vol. 3, pp. 1645-1656.
distribution system are seldom dispatched and controlled by [16] P A Daly, J Morrison," Understanding the potential benefits of
operators, some control means could not adapt to operation distributed generation on power delivery systems," in Proc. Of 2001
mode of steep fluctuation. So planning scheme proposed in Rural Electric Power Conference, pp. A2-1-A2-13.
[17] Viawan F, Sannino A, "Voltage control with distributed generation and
this paper is well suitable to various operational modes.
it impact on losses in LV distribution systems," Power Tech, 2005
It should be noted, the following condition is assumed in IEEE Russia, pp. 1-7.
this paper: DG units have the same performance; the [18] K P Wong, "Recent development and application of evolutionary
meteorological phenomena of installment sites are identical. optimization techniques in power systems, "Proceedings on
international conference on Power System Technology, 2002.
Load changes didn’t take into account.
[19] Kennedy J, Eberhart R C, "Particle swarm optimization," IEEE
conference on Neural Networks (Perth, Australia),IEEE Service Center,
Piscataway, NJ, 1995.
VI. CONCLUSION [20] Mesut E, Baran, Felix F Wu, "Network reconfiguration in distribution
system for loss reduction and load balancing,"IEEE Transactions on
In this paper, optimal location and capacity of distributed Power Delivery, 4 (2), pp. 1401-1407, 1989.
generated is investigated. As the uncertainty of DG output
power, Scenario probability methodology is introduced to
establish synthetic planning model. The proposed method is VIII. BIOGRAPHIES
applied to IEEE-33 bus distribution system. Planning DG Ding Xiaoqun was born in Danyang in China, in October, 1956. He
embedded in distribution system can well adapt various graduated from Southeast University.
He studied and living in Virginia Tech and Univ in the USA for about four
operational modes of DG output. Network energy loss and
years. Recently he mainly engages in artificial intelligence application of
investment cost is optimal in whole planning level year. power system analysis and control, teaching and scientific research in fault
diagnosis of electric equipment and automation of distribution network field.
VII. REFERENCES Wu Jiahong was born in Hubei province, China, in 1979. He graduated
from Nanjing Institute of Technology, Nanjing, China in 2001.
[1] Thomas Ackermann, Goran Andersson, and Lennart Soder, From 2001 to 2006, he worked for Beijing Sifang automation co. ltd.
"Distributed generation: a definition," Electric Power Systems
Currently, He is currently working for the M.Sc degree in Hohai
Research, vol. 57, pp. 195-204, 2001.
University. His research interests are automatic voltage control and
[2] Hasan Hedayati, S A Nabaviniaki, Adei Akvbarimajd, "A method for
distributed generation planning.
placement of DG units in distribution networks," IEEE Trans on Power
Delivery, 23(3), pp. 1620-1628, 2008.
[3] S. A. Papathanassion, "A technical evaluation framework for the
connection of DG to the distribution network," Electric Power systems
research. vol. 77, pp. 24-34, 2007.
[4] EI-Khattam, and M.M.A. Salama, "Distributed generation
technologies, definitions and benefits," Electric Power Systems
Research, vol. 71, pp. 119-128, 2004.
[5] Caisheng Wang, M. Hashem Nehrir, "Analytical Approach for optimal
placement of distributed generation sources in power system,” IEEE
Transaction on Power Systems, 19(4), pp. 2068-2076, 2004.
[6] FAVUZZA S, GRADITI G, IPPOLITO M G, et al, "Optimal electrical
distribution systems reinforcement planning using gas micro turbines
by dynamic ant colony search algorithm, " IEEE Trans on Power
Systems, 22(2), pp. 580-587, 2007.
[7] EL-KHATTAM W, BHATTACHARYA K, HEGAZY Y, et al,
"Optimal investment planning for distributed generation in
competitive electricity market, " IEEE Trans on Power Systems, 19(3),
pp. 1674-1684, 2004.
[8] EL-KHATTAM W, HEGAZY Y, SALAMA M M A, "An integrated
distributed generation optimization model for distribution system
planning, " IEEE Trans on Power Systems, 20(2), pp. 1158-1165, 2005.
[9] R. E. Brown, J. Pan, X. Feng, and K. Koutlev, "Sitting distributed
generation to defer T&D expansion, " IEEE Transmission and
Distribution Conf. and Expo, vol. 2, pp. 622-627, 2001.
[10] Rahman T.K.A, Rahim, Musirin I, "Optimal allocation and sizing of
embedded generators," IEEE Power and Energy conference, Kuala
Lumpur, Malaysia, pp. 288-294, 2004.
[11] Senjyu T, Miyazato Y, Yona, A et al, "Optimal distribution voltage
control and coordination with distributed generation," IEEE Trans on
Power Delivery, 23(2), pp. 1236-1242, 2008.
[12] GAO CI-Wei, CHENG Hao-zhong, Wang Xu, "Electric power network
flexible planning model based on the probability of scene occurrence,"
Proceedings of the CSEE, 24(11), pp.34 -38, 2004.