Beruflich Dokumente
Kultur Dokumente
Abstract: A strategy is presented for the placement of distributed generation (DG) units in the dis-
tribution networks in an uncertain environment. Uncertainties in the system are modelled using
fuzzy numbers. The proposed approach is based on a multi-objective model in which the objectives
are defined as minimisation of monetary cost index (including investment, operation cost of DG
units and cost of losses), technical risks (including risks of voltage and loading constraints violation
because of load uncertainty) and economic risk due to electricity market price uncertainty. The true
Pareto-optimal solutions are found with a multi-objective genetic algorithm and the final solution is
found using a max – min approach. An example is presented to demonstrate the effectiveness of the
proposed methodology.
252 IET Gener. Transm. Distrib., 2008, 2, (2), pp. 252 –260
Table 1: DG placement methods
Rau and Wan [2] minimising of loss, line loading one load level gradient and second-order
and reactive power method
requirement in the network
Kim et al. [3] minimising of loss multi-load level hybrid fuzzy nonlinear goal
programming and GA
Nara et al. [4] minimising of loss uniformly distributed constant tabu search
current with unity power
factor
Griffin et al. [5] minimising of loss peak load level heuristic iterative method
Kim et al. [6] minimising of loss one load level Hereford Ranch algorithm
Celli et al. [7] minimising cost of network peak load with constant growth hybrid e -constraint-based
upgrading, power losses, rate multi-objective programming
energy not supplied and and GA
energy required by the
customers
Willis [8] loss minimisation uniformly distributed load analytical based on 2/3 rule
El-Khattam et al. [9] minimising cost of investment multi-load level heuristic iterative search
and operation of DGs, loss method
and energy required by the
customers
Wang and Nehrir [10] minimising of loss time-varying load analytical approach
Harrison and Wallace [11] maximising DG capacity minimum load level optimal power flow
Popoviv et al. [12] maximising DG capacity multi-load level sensitivity analysis
Keane and O’Malley [13] maximising DG capacity one load level linear programming
Carpinelli et al. [14] minimising cost of losses and peak load with constant growth hybrid e -constraint-based
improvement in voltage rate multi-objective programming
quality and harmonic and GA
distortions
Teng et al. [15] minimising losses and customer average load level GA
interruption costs
Borges and Falcao [16] minimising of losses and one load level GA
guaranteeing acceptable
reliability level and voltage
profile
Acharya et al. [17] minimising of loss peak load level analytical approach
Expression (1), being fuzzy, does not have a simple ‘true’ line segment (or substation) is presented as the crisp value
or ‘false’ value. It is violated only with a certain degree of (Pmax). Same concept as that used for the voltage constraint
possibility, defined as can be used for the expression of loading constraint in the
fuzzy domain [20] with
Aol þ Aor
SVk ¼ 100 (%) (2)
Ai þ Aol þ Aor
P~ k
~ Pmax (3)
where Ai is the area under the membership function between
the upper and lower voltage limits. Also, Aol and Aor are the Similar to the voltage constraint, possibility degree of viola-
areas under the membership function left of the Vmin and tion for the loading constraint can be defined as
right of the Vmax , respectively (Fig. 1b). Basically, this
fuzzy constraint means that voltage constraint will be vio- Ar
lated with the possibility SVk . Hence, by changing the par- S Pk ¼ 100 (%) (4)
ameters of TFN V~ k , the ratio between areas Ai and Al þ Ar
(Aol þ Aor) also changes, affecting the possibility SVk as
well. where Al and Ar are the areas under the membership func-
By running the fuzzy load flow in a distribution network tion left and right of the loading limit Pmax , respectively
for a given condition, the result of power flow through the (Fig. 1c). In general, larger values of SVk and SPk mean
line segments and substations are also fuzzy numbers. In greater risk in violation of voltage constraint at node k
Fig. 1c, this power flow in the line segment (or substation) and loading constraints in the line segment (or substation)
k is shown as TFN (P~ k ), whereas the loading limit in this k, respectively.
IET Gener. Transm. Distrib., Vol. 2, No. 2, March 2008 253
Technical risk items due to uncertainties in load
Economic risk
X
T X
NN
þ 8760 mt PDGi OCDG
t¼1 i¼1
X
T
8760 mt (P~ Loss0 P~ Loss )C~ T (5)
t¼1
3.3 Economic risk objective function In this model, constraint on DG operation, in which the
power generated from DG must be less than the DG
Payment towards purchased power is proportional to the capacity, is modelled by (18). In this equation, PFDG is
multiple of the power and the electricity market price. the power factor of generated power of DG. Also, (19)
Thus, the economic risk due to the increase in the electricity shows the constraint on DG installation in candidate sites,
market price can be expressed as the possibility degree of which is applicable in cases where, due to practical or geo-
the following inequality graphical limitations, DG installation is not feasible for
more than a certain capacity.
C~ E C~ E0 (12) In a multi-objective minimisation problem, a feasible sol-
ution X is said to dominate another feasible solution Y, if
X
NS X
NN and only if (i) the solution X is not worse than Y over all
C~ E ¼ P~ Si C~ T þ PDGi OCDG (13) objectives and (ii) the solution X is strictly better than Y
i¼1 j¼1 in at least one objective. A solution is said to be Pareto
optimal if it is not dominated by any other solution in the
X
NS
C~ E0 ¼ P~ S0i CT max (14) solution space. A Pareto-optimal solution cannot be
i¼1 improved with respect to any objective without worsening
at least one other objective. The set of all feasible non-
where P~ S0i and P~ Si are the powers purchased by the DisCo dominated solutions is referred to as the Pareto-optimal set.
from the main grid through the ith substation as TFN (in There are different algorithms for the solution of a multi-
MW) before and after DG placement, respectively. PDGj objective problem. A multi-objective optimisation algor-
is the power generated from the jth DG as crisp number ithm must not only guide the search towards the global
(in MW). Pareto-optimal solutions, but also maintain the population
In (12), costs of meeting customer energy demand in two diversity in the Pareto-optimal set. The problem considered
here is a combinatorial problem and the objective functions
cases with and without the presence of DGs in the distri-
are nonlinear with fuzzy numbers. Thus, classical optimis-
bution network are compared. With respect to graphical rep-
ation methods cannot be implemented easily in this case.
resentation of C~ E and C~ E0 in Fig. 2, the possibility degree of
In this paper, the NSGA-II [18], which incorporates the
fuzzy inequality (12) is defined as
concept of Pareto optimality into its search algorithms and
can find optimal trade-offs among the multiple conflicting
Br objectives simultaneously, has been developed and
SCE ¼ 100 (%) (15)
Br þ Bl implemented.
Thus, the economic risk objective function is 4 Implementation of NSGA-II for DG placement
problem
Min fE ¼ SCE (16)
4.1 Decision variables coding
It is observed from the concept of this objective that DG In a DG placement problem, the decision variables are the
installation in a distribution network enables DisCo to sizes and generated power of DG units at specified
meet part of its customer energy demand by the operation locations. Because of the difficulties of binary coding,
of DG units with crisp operation costs and purchase less when dealing with continuous search space with large
power from the main grid for a uncertain price. In this dimension, in the proposed approach the decision variables
way, DG installation and operation can reduce the economic are coded using real numbers as
risk.
X ¼ {PDG1 , PDG2 , PDG3 , . . . , PDGNN } (20)
3.4 Multi-objective model of DG placement The ith gene in this chromosome demonstrates the power
generated by DG at the ith candidate location within its
In this paper, DG placement is modelled as a multi- lower limit ai ¼ 0 and upper limit bi ¼ PFDG . CDG max i .
objective model with the three objective functions
4.2 Computational procedure
Step 4: Objective evaluation. In this step, the values of 4.3 Selection of final solution
objective functions for each member of current population
are evaluated. To select the final multi-objective solution after analysing
the set of Pareto-optimal solutions, planners can select the
Step 5: Non-dominated ranking. The members of popu-
final DG placement solution, considering the most satisfac-
lation are ranked based on non-domination with rank
tory values of the three objectives and according to their
identifier.
experience and professional point of view. In this paper, a
The monetary objective function value is a TFN and has
max – min approach is used to select the best (final) multi-
to be compared and ranked in this step. The comparison of
objective DG placement solution. Each solution in the set
this fuzzy value is carried out using the defuzzified value. of Pareto-optimal solutions has an associated vector of
Defuzzification methods are based on heuristic ideas. The values (f~ Mk , fTk , fEk ) that can be normalised using the fol-
choice of an appropriate defuzzification method depends lowing expressions
on the application under consideration and can be based
on an axiomatic and/or an empirical justification [22, 23]. fM max R( f~ Mk )
In this paper, the defuzzification of the monetary objective fMnk ¼ (26)
fM max fM min
function value is done using removal function R(~a) that for a
TFN, a~ ¼ (aL , aM , aR ) is shown as [23] fT max fTk
fTnk ¼ (27)
R(~a) ¼ (aL þ 2aM þ aR )=4 (24) fT max fT min
fE max fEk
Step 6: Crowding distance computation. Since the individ- fEnk ¼ (28)
uals are selected based on the rank identifier and crowding fT max fT min
distance, all the individuals in the population are assigned a where fM max , fT max and fE max are the maximum values
crowding distance. The crowding distance computation obtained for the monetary objective function [defuzzified
requires sorting of the population according to each objec- value obtained using (24)], technical risk objective function
tive function value in their ascending order of magnitude. and the economic risk objective function, respectively, and
Thereafter, for each objective function, the solutions are fM min , fT min and fE min the minimum values obtained. Note
assigned a distance value equal to the absolute difference that the result of this normalisation shows the level of con-
in the function values of two adjacent solutions in the tentment for each objective function. Afterwards, a max –
same front. This calculation is continued with other objec- min approach, shown in (29), is applied to select the best
tive functions. The overall crowding distance value is calcu- (final) multi-objective DG placement solution, that is, the
lated as the sum of individual distance value corresponding most suitable solution using the above-mentioned approach
to each objective function [18]. Crowding distance is [21, 23]
assigned front wise and comparing the crowding distance
between two individuals in different fronts is meaningless.
Mathematically max min{fMnk , fTnk , fEnk } (29)
k
Xm
fj (Xiþ1 ) fj (Xi1 ) 5 Application study
cd(Xi ) ¼ , i [ F r (25)
j¼1
fjmax fjmin
The proposed methodology for DG placement has been
where cd(Xi) is the overall crowding distance of solution Xi , implemented in the MATLAB environment and tested on
256 IET Gener. Transm. Distrib., Vol. 2, No. 2, March 2008
maximum allowable loading of substation is considered as
52.5 MV A.
Technical characteristics of conductors used in line seg-
ments and future load data are given in Tables 2 and 3,
respectively. All network nodes except substation node
are considered as candidate sites for DG placement.
Table 2: Technical characteristics of conductors and future load data of the distribution network
Power factor for all loads is 0.95. The candidate DGs have
sizes of multiples of 1 MV A (maximum 4 MW) with power
generation at load factor of 0.9. Cost data are listed in
Table 4. In this table, electricity market price (C~ T ) is a
TFN. Allowable voltage deviation is considered as
DV ¼ +5%.
The proposed NSGA-II has been executed with the fol-
lowing parameters:
ohmic power losses (as TFN) P̃Loss MW (1.314, 1.516, 1.735) (0.562, 0.684, 0.823)
substation loading (as TFN) P̃S MV A (46.699, 51.920, 57.151) (33.326, 38.492, 43.669)
risk of overloading in substation SPS -MAX % 39.57 0
risk of overloading in line SPL-MAX % 100 2.58
segments
risk of over/under voltage in SVD-MAX % 100 0
load nodes
monetary objective function f̃M $ 0 4.854 107
(removal value)
technical risk objective function fT % 100 2.58
economical risk objective fE % 16.74 7.50
function
reduced to 2.58%. Economic risk, as defined in Section 3.3, Also the application of proposed method for DG place-
changed from 16.74% in the initial condition to 7.50% after ment with stochastic generation is under study.
DG installation.
Also, it should be noted that in the proposed multi-
objective risk-based DG placement, the Pareto-optimal 7 References
solutions obtained give the information about the existing
trade-off among the three objectives. Each solution of this 1 Ackermann, T., Andersson, G., and Soder, L.: ‘Distributed
set is useful and shows a strategy for DG placement. If the generation: a definition’, Electr. Power Syst. Res., 2001, 57, (3),
pp. 195–204
planners prefer a DG placement strategy with lowest 2 Rau, N.S., and Wan, Y.-H.: ‘Optimum location of resources in
economic risk, they will select the solution marked in distributed planning’, IEEE Trans. Power Syst., 1994, 9, (4),
Fig. 4 as ‘solution with minimum economic risk’, in pp. 2014–2020
which the economic risk is eliminated. This solution is 3 Kim, K.-H., Lee, Y.-J., Rhee, S.-B. et al.: ‘Dispersed generator
shown in Fig. 5b. If the planner want to have a placement using fuzzy-GA in distribution systems’. Proc. IEEE
Power Engineering Society Summer Meeting, Chicago, USA,
maximum 12% for economic risk with the minimum 21–25 July 2002, pp. 1148– 1153
cost, they will select the solution marked in Fig. 4 as ‘sol- 4 Nara, K., Hayashi, Y., Ikeda, K. et al.: ‘Application of tabu search to
ution with maximum 12% economic risk’, which is rep- optimal placement of distributed generators’. Proc. IEEE Power
resented in Fig. 5c. In this solution, the technical risk is Engineering Society Winter Meeting, Columbus, USA, 28 January–
1 February 2001, pp. 918– 923
7.7%. 5 Griffin, T., Tomsovic, K., Secrest, D. et al.: ‘Placement of dispersed
There are many other DG placement strategies in the generation systems for reduced losses’. Proc. 33rd Int. Conf. System
Pareto-optimal set, which can be selected by the planners Sciences, Hawaii, 4– 7 January 2000, pp. 1446–1454
taking into consideration their experience or the requisites 6 Kim, J.O., Nam, S.W., Park, S.K. et al.: ‘Dispersed generation
of the distribution networks. planning using improved Hereford Ranch algorithm’, Electr. Power
Syst. Res., 1998, 47, (1), pp. 47–55
7 Celli, G., Ghaiani, E., Mocci, S. et al.: ‘A multiobjective evolutionary
6 Conclusions algorithm for the sizing and sitting of distributed generation’, IEEE
Trans. Power Syst., 2005, 20, (2), pp. 750–757
A new strategy for DG placement problem is presented in 8 Willis, H.L.: ‘Analytical methods and rules of thumb for modeling
DG-distribution interaction’. Proc. IEEE Power Engineering
this paper. The model considers a fuzzy explicit represen- Society Summer Meeting, Seattle, USA, 16–20 July 2000,
tation of the uncertainties associated with the future load, pp. 1643–1644
as well as a fuzzy representation of the uncertainties 9 El-Khattam, W., Bhattacharya, K., Hegazy, Y. et al.: ‘Optimal
associated with the power flow in the feeders and substa- investment planning for distributed generation in a competitive
tions, the network node voltages and the electricity market electricity market’, IEEE Trans. Power Syst., 2004, 19, (3),
pp. 1674–1684
prices. This model provides a simultaneous minimisation 10 Wang, C., and Nehrir, M.H.: ‘Analytical approaches for optimal
of the economic cost and technical and economic risks. placement of distributed generation sources in power systems’,
A specialised NSGA-II has been proposed as a solution IEEE Trans. Power Syst., 2004, 19, (4), pp. 2068– 2076
algorithm. The algorithm determines the set of 11 Harrison, G., and Wallace, A.: ‘Optimal power flow evaluation of
distribution network capacity for the connection of distributed
Pareto-optimal multi-objective DG placement solutions. generation’, IEE Proc., Gener. Trans. Distrib., 2005, 152, (1), pp. 115–122
Planners can select the most satisfactory Pareto-optimal 12 Popovic, D.H., Greatbanks, J.A., Begovic, M. et al.: ‘Placement of
solution on the basis of their experience and professional distributed generators and reclosers for distribution network security
point of view. This feature illustrates the possibilities for and reliability’, Int. J. Power Energy Syst., 2005, 27, (5–6), pp. 398–408
practical application and the flexibility of the proposed 13 Keane, A., and O’Malley, M.: ‘Optimal allocation of embedded
generation on distribution networks’, IEEE Trans. Power Syst.,
methodology. In this paper, a max – min approach has 2005, 20, (3), pp. 1640–1646
been proposed to select the best Pareto-optimal DG place- 14 Carpinelli, G., Celli, G., Mocci, S. et al.: ‘Optimisation of embedded
ment solution. The new DG placement algorithm has been generation sizing and sitting by using a double trade-off method’, IEE
tested in a typical distribution network. The obtained Proc., Gener. Trans. Distrib., 2005, 152, (4), pp. 503–513
15 Teng, J.-H., Luor, T.-S., and Liu, Y.-H.: ‘Strategic distributed
results show that the suggested DG placement model is generator placements for service reliability improvement’. Proc.
a powerful decision-making tool for risk management in IEEE Power Engineering Society Summer Meeting, Chicago, USA,
distribution networks with DG installation and operation. 21–25 July 2002, pp. 719–724