A Clustering based Wind Farm Collector

System Cable Layout Design

S. Dutta, Student Member, IEEE, and T. J. Overbye, Fellow, IEEE

wind farm losses. This paper proposes a cable layout design

Abstract— The goal of achieving 20% wind power penetration by methodology that reduces active power losses in the collector
2030 in the US has stimulated the installation of large scale wind system significantly compared to the conventional radial
farms in recent years, both on-shore and off-shore. Collector system configuration while improving the reliability of the
systems consolidate the power generated by turbine units system.
distributed over the geographical area of the wind farm to a
Wind turbine locations and point of interconnection to
substation from where the generated power is transmitted to the
electric grid. Design of a wind farm collector system must take the electric grid are the primary factors in the design and
into consideration the economics and reliability of operation. layout of feeder topology. Other factors include terrain,
Most modern day large scale wind farms consist of hundreds of reliability, landowner requirements, power losses, economics,
wind turbines and are generally electrically connected in a radial and climate of the location [1]. Since the point of
feeder cable configuration or daisy chains. While these interconnection is typically located at the wind farm
configurations are generally accepted as convention, not much substation, for this work, only the feeder topology from
research has been done to analyze other cable layout turbine locations to the substation has been considered.
configurations. Collector system electrical power losses impact the
This paper proposes a clustering based algorithm for cable economic evaluation of wind farm operation. Investments for
layout of a large scale wind power plant. Comparison of the loss reduction can generate substantial returns in the long-term
proposed method with the radial feeder cable configuration
[2]. Intuitively, to lower power losses, the conductor lengths
shows that real power losses in collector system are lowered and
greater reliability is achieved with the proposed design. An should be shorter and area of cross-section should be larger.
economic analysis has also been done to compare the cost of However, shortening conductor lengths may not always be
generated energy associated with the proposed design and the possible due to constraints of land owners, terrain, access
conventional configuration. roads, and point of interconnection. In addition, the cable
sizing is typically dictated by the costs, thermal resistivity of
the soil, ampacity requirements, short-circuit withstand
Index Terms— cable layout, clustering algorithm, power loss, requirements, and conductor arrangement in the trench.
quality threshold clustering, wind farm collector system. Collector systems can be structured in different
configurations with varying levels of collector system
reliability. Typical collector system configurations are:
I. INTRODUCTION • Loop system- This system provides a redundant path for the

U NLIKE traditional power generating plants which are generator output for each turbine by establishing a looped
built around a few high rating generating units within a circuit between the wind turbines. In the event of cable failure,
single location, wind farms aggregate the power generated by the loop can be opened and the full output of the wind farm
a multiplicity of small wind powered generators spread out can be maintained.
over a large area. The energy generated by each unit is • Single string- This system places all of the wind turbines on
collected and channeled to a substation by means of a network a single series circuit. In the event of a cable failure, the wind
of medium-voltage cables called a collector system. Due to turbines located beyond the faulted cable would not be
available until the cable is repaired.
the large separation between the wind turbines and the large
area of wind farms, typical cable collector systems can • Multiple string- This system distributes the wind turbines
measure over a hundred three-phase circuit miles. Hence, the over several series circuits and permits the use of lower rated
reliability of the entire wind farm is strongly impacted by the equipment. Similar to the single string configuration, in the
event of cable failure the wind turbines beyond the faulted
reliability of the collector system. The large expanse and
cable will not be available until the cable is repaired.
medium voltage of the collector system also results in the
Radial feeder or string configurations are commonly
collector system representing the largest portion of the total
used in the US. However, these configurations compromise
system reliability since a fault in a given collector circuit cable
The authors would like to acknowledge support provided by the Stanford
will result in an outage of all wind turbines connected to that
University Global Climate and Energy Project (GCEP) and the Power particular circuit.
Systems Engineering Research Center (PSERC). Feeder strings may be underground or overhead. The
The authors are with the Department of Electrical and Computer number of wind turbines on a feeder string is limited by
Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801 conductor ampacity. Underground feeders are generally
USA (e-mail: [email protected]; [email protected]).
limited to 25 to 30 MW per string due to soil thermal

978-1-4244-8052-4/11/$26.00 ©2011 IEEE

conditions and available cable sizes. Overhead conductors can laid out from each of the turbine locations of that group to the
carry 40 to 50 MW per string but have higher losses first level cluster representative point. The cable size is
associated. selected from the ampacity required for the cable to carry
Most wind farms today connect the wind turbines in a power from the turbine at medium level collector system
radial feeder configuration or daisy chains. While these voltage.
configurations are generally accepted as convention, not much In the next level, the first level cluster representative points
research has been done to analyze other cable layout are clustered to compute second level cluster groups and
configurations. Although available literature documents work second level cluster representative points similar to the first
done on layouts of large scale wind farms such as evaluation level. This process is followed till the number of points to be
of AC and/or DC collection systems in varying voltage levels clustered reaches a prespecified tolerance level or the cable
[3], most of these works assume turbines connected in a daisy ampacity warranted to carry the power exceeds current
chain. The present paper proposes a clustering based cable carrying capabilities of available cable sizes.
layout configuration. Clustering analysis has been used in In this work, three different cable layout configurations
earlier research works on wind turbines, but has not been used have been compared. These are (i) a pure clustering based
for collector system design. Most clustering algorithms were method, (ii) a conventional radial system, and (iii) a mixed
used to group wind turbines based on power output for system which combines clustering and radial systems. The
controller system design and development of power comparison in between the three different systems has been
forecasting algorithms [4]. made from the perspectives of overall cost, system reliability,
The remaining paper is organized as follows: Section II and system real power losses. It has been assumed that
discusses some of the mathematical considerations of the Aluminum strand conductors of a few sizes are available as
addressed problem. Section III presents the proposed described in section V. Approximate cost and ac resistance
methodology. Section IV describes the quality threshold values used for this work are also mentioned.
clustering algorithm. Section V describes the sizes and
properties of available cables for the collector system in this
paper. Section VI focuses on the procedure for cost analysis. IV. QUALITY THRESHOLD CLUSTERING
Section VII presents the results and discussions. Finally, The QT (Quality Threshold) Clustering algorithm [6] was
conclusions have been made in Section VIII. developed to group genes into high quality clusters. Quality is
ensured by finding clusters whose diameters do not exceed a
given user-defined diameter threshold. This method prevents
II. MATHEMATICAL CONSIDERATIONS dissimilar genes (wind turbine locations in this work) from
The problem of cable layout design for a wind farm being forced under the same cluster and ensures that only
collector system can be considered as finding a tree to meet good quality clusters will be formed. The goal of QT
required design characteristics in a graph G = (V, E), where V clustering is to form large clusters with similar expression
represents the set of vertices or locations of wind turbines and pattern, and to ensure a quality guarantee for each cluster.
the substation, and E represents the set of edges connecting the Quality is defined by the cluster diameter and the minimum
vertices. Given the set of vertices, there can be numerous number of points contained in each cluster. Compared to K-
possible trees satisfying specific design constraints. Hence, means, another popular clustering algorithm, Quality-
finding an optimal tree can be extremely difficult especially if Threshold clustering has been used in this work since in QT
the problem allows the addition of intermediate vertices and clustering there is no need to specify the number of clusters, as
edges to the graph in order to improve the value of the required in K-means. Also, it is not necessary that all points
objective function. For instance, given a large number of wind need to be clustered and returns the same result when run
turbine locations, the problem of minimizing the total length several times. So although QT is computationally more
of cables is NP-hard [5]. Hence, in this work, instead of expensive than K-means, this was selected as a clustering
addressing the optimality of the proposed cable layout, an algorithm.
effort has been made to develop a cable layout algorithm that The QT algorithm is as follows [7]:
improves the collector system by lowering power losses and 1. A random point is chosen from the list of points to be
improving the reliability compared to the conventional clustered.
configuration. 2. The algorithm determines which point has the greatest
similarity (closest Euclidean distance) to this point. If the
distance is less than the quality threshold distance which is
III. METHODOLOGY pre-specified, then these two points are clustered together.
3. Other points are similarly added to this cluster. This process
This method assumes that locations for the wind turbines
continues until no point can be added to this first candidate
considering local wind pattern, wake effect, etc are already
available. The cable layout method presented here is based on cluster without surpassing the diameter threshold.
applying quality-threshold clustering on wind turbine 4. A second candidate point is chosen.
locations in multiple levels. 5. The algorithm determines which point has the greatest
At the first level, the wind turbine locations are clustered similarity (closest) to this second point. All points in the list
into a number of groups. The turbine location in each group of points are available for consideration to the second
that is closest in distance to the substation is called the first candidate cluster.
level cluster representative point for that group. Cables are
6. Other points from the list of points that minimize the VI. ANALYSIS OF COSTS
increase in cluster diameter are iteratively added to the A levelized cost estimate for energy delivered by the
second candidate cluster. The process continues until no wind farm has been found by computing the ratio of annual
point can be added to this second candidate cluster without costs and the annual energy produced. The annual cost
surpassing the diameter threshold. incurred is the sum of annual loan payments, annual return on
7. The algorithm iterates through all points on the selected list equity and operation and maintenance costs. The annual
of points and forms a candidate cluster with reference to energy generated is found by the product of the nameplate
each point. In other words, there will be as many candidate capacity of the wind farm reduced by the collector system
clusters as there are points in the list. Once a candidate losses, number of hours in a year, and the capacity factor of
cluster is formed for each point, all candidate clusters below the wind farm. The capital cost of the wind farm project is
the user specified minimum size are removed from computed by adding the cost of the turbine units
consideration. ($1million/1MW turbine), trenching costs ($15/ft), and cable
8. The largest remaining candidate cluster, with the user- costs and then multiplying with a factor of 1.05 to take into
specified minimal number of points, is selected and retained account other components of the capital cost such as site
as a QT cluster. The points within this cluster are now preparation, grid connections, project development, and
removed from consideration. All remaining points will be feasibility study. Some cost figures for different cable sizes for
used for the next round of QT cluster formation. this analysis are as follows:
9. The entire process (step 1 to 9) is repeated until the largest
remaining candidate cluster has fewer than the user- TABLE II
specified number of points. ACSR CABLE COSTS
10. The result is a set of non-overlapping QT clusters that Al Strand Conductor Size Cost ($/ft)
meet quality threshold for both size, with respect to number 1/0 5
4/0 10
of points, and similarity, with respect to maximum 500 kcmil 12
allowable diameter. 750 kcmil 26
11. Points that do not belong in any clusters will be grouped 1000 kcmil 38
under the “unclassified” group.
The capital cost is spread out over the projected lifetime
using a capital recovery factor (CRF) and adding an estimate
V. AVAILABLE CONDUCTOR SIZES AND PROPERTIES of annual operation and maintenance costs. It is assumed that
It is assumed that ACSR (Aluminum Conductor Steel the wind project is financed by a loan of 75% of the capital
Reinforced) cables are used for the collector system cable cost with interest rate of 7% and loan term of 20years. The
layout. Further, it is assumed that only the cable sizes annual payment on the loan is given by [11]:
mentioned in Table I are available. A P P. CRF i, n (1)
The cable ampacity is based on conditions [8] that cables
Where A is the annual payment in $/yr, P is the principal
are installed in sand with minimum cover of 36 inches, load
borrowed in $, i is the interest rate, n is the loan term in years.
factor is 100%, and maximum ambient earth temperature is 20
The operation and maintenance costs are assumed to be 3% of
deg C. The dc resistances of the different conductors were
the capital cost and include parts and labor, insurance,
obtained from [8], [9]. The ac resistance of a conductor is the
contingencies, land lease, property taxes, transmission line
dc resistance increased by a skin effect factor and a proximity
maintenance, general and miscellaneous costs. Furthermore, a
factor: R R 1 γ γ , where R is dc resistance 15% return on equity is taken.
at desired operating temperature, γ is the skin effect factor,
and γ is the proximity effect factor. The ac resistances are
VII. RESULTS AND DISCUSSIONS
computed from the dc resistances according to the method
described in [10] and are tabulated below. For this study, a 22MW wind farm with twenty two 1MW
wind turbine units has been considered. Underground cables
TABLE I interconnect turbines and the substation in the 34.5kV medium
ACSR CABLE SIZES AND PROPERTIES voltage collector system. Assuming the units generate power
Al Strand Continuous DC Resistance at AC Resistance at
Conductor Size Ampacity 25 deg C 25 deg C
at a lagging power factor of 0.8, the current is projected to be a
(Amps) (ohms/1000 ft) (ohms/1000ft) maximum of 20.92 A per unit at 34.5kV.
[9] . .
√
1/0 150 0.1672 0.1672
4/0 211 0.0836 0.0836
Optimal wind turbine placement considering effects of local
500 kcmil 332 0.0354 0.0361 wind patterns, land topography, wind turbine wake effects is a
750 kcmil 405 0.0236 0.0248 widely researched area. Therefore, for the purpose of this
1000 kcmil 462 0.0176 0.0193 work, it has been assumed that optimal turbine locations in the
wind farm are already available. This assumption is justified
by the availability of abundant literature in this area. This
work does not consider issues of laying cables such as land
unavailability within the wind farm area or trenching
restrictions.
 It has been assumed that the average wind speed at the to achieve specific design targets and meet constraints such as
wind farm location is 8.5m/s. Assuming Rayleigh distributed maximum number of turbines in a cluster.
wind speeds, capacity factor of the turbines can be computed
using the following relation [11]: 10
PR
CF 0.087V (2) 9
D
Where, V is the average wind speed in m/s, PR is the rated 8

power in KW of a wind turbine, D is rotor diameter in m. With 7

rated power of each turbine unit taken as 1MW, rotor diameter

Y-axis: 1 unit=1000ft
6
52m, and average wind speeds of 8.5m/s, the capacity factor is
approximately 37%. In figure 1, the locations of the 22 wind 5

turbines and the substation are shown. The wind turbines are 4
shown as dots and the substation location is indicated by a 3
filled square. The figures in this paper have been scaled so that
each unit on X and Y axes correspond to 1000ft. Thus, the 2
First level cluster
minimum distance between any two turbine locations is 1 representative points
1000ft. 0
The locations of the wind turbine units are given by the
following set of points in cartesian coordinates: -1
-1 0 1 2 3 4 5 6 7 8 9 10
{(0,0), (1,0), (2,0), (0,1), (1,1), (2,1), (0,2), (1,2), (2,2), (3,2), X-axis: 1 unit=1000ft
(0,3), (1,3), (3,3), (1,4), (5,5), (6,4), (6,6), (7,5), (7,6), (7,7), Fig. 2. First level clustering
(8,6), (8,7)}
Figure 2 shows the two clusters formed containing
10
respectively 14 and 8 wind turbines. The first level cluster
9 representative points, shown by squared dots in each cluster
8
are the wind turbine locations in each cluster which are closest
to the substation. Cables are laid out from turbine locations in
7 a cluster to the cluster representative point. Since each of these
first level cluster cables must carry a maximum of 20.92A, 1/0
Y-axis: 1 unit= 1000ft

6
conductor sized cables can be used from the available
5
conductor sizes. The total length of 1/0 conductor sized cables
4 is 47.3701units = 47370.1ft.
3 Wind Maximum losses associated with the 1/0 conductor sized
turbine cables is 10398.8W.
2
In the second level, the quality threshold distance is taken
1 as 4units = 4000ft. Figure 3 shows the cable layouts after the
Substation
0 second level clusters are formed. After the second level
clustering, the two cluster representative points from the
-1
-1 0 1 2 3 4 5 6 7 8 9 10 previous level get clustered. The second level cluster
X-axis: 1 unit=1000ft representative point is marked as a diamond around a black
Fig. 1. Locations of 22 wind turbines (dots) and one collector substation dot.
(square)
10
The collector system has been studied for three different 9
cable layout systems. The first method studied is the proposed
quality threshold based clustering method, the second method 8

considered is the conventional radial method, and the third is a 7

combined clustering-radial method.
Y-axis: 1 unit= 1000ft

5
A. Cluster based method
4
The proposed cluster based method is used to lay out cables
in the wind farm. At the first level of clustering, a quality 3

threshold distance of 2.5units = 2500ft is used. The choice of 2

Second level cluster
quality threshold distance for each stage of clustering is by 1 representative point
trial and error. A good initial guess for the first stage
clustering can be 2 to 3 times the minimum distance in 0

between turbines. For every subsequent stage the initial guess -1

-1 0 1 2 3 4 5 6 7 8 9 10
can be double the quality threshold distance for the previous X-axis: 1 unit=1000ft
stage. The initial guess for the threshold distance is modified
Fig. 3. Second level clustering
 Only a single cable is laid out in this level which carries 10

the power from the eight-turbine cluster and hence a total 9

maximum current of 167.36A. Thus a 4/0 conductor sized 8
cable can be used. Length of the required cable is 3.6056units
= 3605.6ft. The power losses associated with this cable is 7

25328.4W.

Y -axis: 1 unit=1000ft
6
In the last level, a cable carries power from the second 5
level cluster representative point to the substation. This cable
4
carries power from all the turbines in the wind farm and
carries a maximum current of 460.24A. The cable used for this 3

is one with 1000kcmil sized conductors. The length of the 2

required cable is 2.8284units = 2828.4ft and losses associated 1
equal 34688.7W. Figure 4 shows the cables laid out at the end
of the third and final level of quality threshold clustering. 0

-1
10 -1 0 1 2 3 4 5 6 7 8 9 10
X-axis: 1 unit=1000ft
9
Fig. 5. Radial system cable layout
8

7
Assuming all the wind turbine units generate their rated
power and the currents are all in phase, an equivalencing
Y-axis: 1 unit=1000ft

6
process described in [12] is used to compute the currents in
5 each cable.
4
The cables required in this method are 24376ft of 1/0
conductor sized cables, 1000ft of 4/0 conductor sized cables,
3
and 3000ft of 500kcmil conductor sized cables.
2

1 The total loss in the cables is 123.9kW.

This is 0.56% of the installed wind farm capacity.
0
Total cable costs = $167880
-1
-1 0 1 2 3 4 5 6 7 8 9 10
Total trenching costs = $425640
X-axis: 1 unit=1000ft Turbine costs = $22,000,000
Fig. 4. Third and final level clustering
Capital costs = $22,593,520 * 1.05 = $23,723,196
Cost of electricity = 4.753cents/KWh
The total cable length required in this method is 161412.3ft.
C. Combined Cluster-Radial system
The total loss in the cables is 70415.97W=70.42kW.
This method uses a combined quality threshold clustering
This is 0.32% of the installed wind farm capacity.
and radial system for the collector system. The cable layout is
Total cable costs = $380,385.7
shown in figure 6. The first level uses the proposed method for
Trenching costs = $807,061.5
clustering the wind turbines. This is followed by a radial
Turbine costs = $22,000,000
interconnection.
Capital costs = $23,187,447.2 * 1.05 = $24,346,819.56
Cost of electricity = 4.741cents/KWh 10

B. Radial System method 8

A conventional method of connecting wind turbines in a 7
wind farm is the radial system or the daisy chain system. The
Y-axis: 1 unit=1000ft

6
radial system can be single or multiple string system. Each
string carries power from one or more wind turbine units to a 5

higher rating feeder cable which carries the power to the 4

substation. Figure 5 shows a possible radial system cable 3
layout configuration for the wind farm considered in this
2
work.
1

-1
-1 0 1 2 3 4 5 6 7 8 9 10
X-axis: 1 unit=1000ft

Fig. 6. Mixed layout configuration

 and the combined cluster-radial method, cable failure results
The total cable length required in this method is 47370ft of 1/0 in a smaller number of turbines going offline. This indicates
conductor sized cables, and 6943ft of 4/0 conductor sized that the proposed method also improves the reliability of the
cables. wind farm collector system.

The total loss in the cables is 99.9kW. REFERENCES

This is 0.46% of the installed wind farm capacity.
Total cable costs = $306,280 [1] E. H. Camm, M. R. Behnke, O. Bolado, M. Bollen, M. Bradt, C. Brooks,
Total trenching costs = $814,695 W. Dilling, M. Edds, W. J. Hejdak, D. Houseman, S. Klein, F. Li, J. Li, P.
Turbine costs = $22,000,000 Maibach, T. Nicolai, J. Patino, S. V. Pasupulati, N. Samaan, S. Saylors, T.
Siebert, T. Smith, M. Starke, R. Walling, “Wind power plant collector
Capital costs = $23,120,975 * 1.05 = $24,277,023.75
system design considerations,” Proc. 2009 IEEE Power and Energy
Cost of electricity = 4.748cents/KWh Society General Meeting, Calgary, Canada, July 2009.
[2] E. H. Camm, M. R. Behnke, O. Bolado, M. Bollen, M. Bradt, C. Brooks,
To summarize the results, the cable layout with the W. Dilling, M. Edds, W. J. Hejdak, D. Houseman, S. Klein, F. Li, J. Li, P.
Maibach, T. Nicolai, J. Patiño, S. V. Pasupulati, N. Samaan, S. Saylors, T.
proposed cluster based algorithm yielded lower power losses Siebert, T. Smith, M. Starke, R. Walling, “Wind power plant substation
compared to the pure radial cable layout configuration. Power and collector system redundancy, reliability, and economics,” Proc. 2009
loss in the combined cluster-radial system was in between the IEEE Power and Energy Society General Meeting, Calgary, Canada, July
pure cluster based system and the pure radial system. 2009.
[3] S. Lundberg, “Evaluation of wind farm layouts,” Available online:
From Figure 4, it can be seen that apart from the cables laid
http://www.elkraft.ntnu.no/norpie/10956873/Final%20Papers/009%20-
out in the second and third level clustering, all the first level %20Eval_WF_lay_009.pdf
cables carry power from one turbine each. This reduces the [4] Y. Ma, J. N. Jiang, and T. Runolfsson, “Cluster analysis of wind turbines
expected number of turbines lost in the event of a cable fault of large wind farm,” IEEE Explore, 2009. Available online:
http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=04840148
in the pure cluster based system. Figure 5 shows that in the
[5] M. Hazewinkel, “Steiner tree problem”, Encyclopaedia of Mathematics,
pure radial system there is a probability of losing greater Springer, ISBN 978-1556080104, 2001. Available at
number of turbines in case of a fault since a greater number of http://eom.springer.de/s/s110270.htm
cables carry power from multiple turbine units. This indicates [6] L. J. Heyer, S. Kruglyak, and S. Yooseph, “Exploring expression data:
identification and analysis of coexpressed genes,” Genome Research, vol.
that the pure cluster based system has higher reliability
9, Nov. 1999, pp. 1106-15.
compared to the pure radial system. The figure 6 shows that [7] QT (Quality Threshold) Clustering, [email protected], 2005.
the combined cluster-radial system replaces the cable laid out Available online:
in the final level of the clustering algorithm carrying power http://www.chem.agilent.com/cag/bsp/products/gsgx/Downloads/pdf/qt_c
lustering.pdf
from all the 22 turbines with two cables carrying power from [8] Maine Mountain Power, LLC, “Preliminary engineering for the Black
14 and 8 turbines each. Hence it provides even greater Nubble wind farm 34.5 kV collector system and 115 kV interconnection
reliability compared to the pure cluster based system. facility 54 MW facility,” June 18, 2007. Available online at:
The costs of generated electricity from all three systems are http://www.maine.gov/doc/lurc/projects/redingtonrevised/Documents/Sec
tion01_Development_Description/Development_Electric/E_Pro_Reports/
comparable, with highest costs for the radial system and
5.1_Electrical_Power_Line_Report.pdf
lowest for the proposed cluster based system. Because of the [9] Kerite Engineering Catalog. Available online:
lower power losses in the cluster system layout, the annual http://www.kerite.com/pdf/Kerite%20Engineering%20Catalog.pdf.
energy generated by it exceeds the same for the radial [10] J. H. Neher and M. H. McGrath, “The Calculation of the temperature rise
and load capability of cable systems,” AIEE Transactions, vol. 76, Oct.
configuration. In addition, although the capital costs are higher
1957, pp. 752–64.
for the proposed cluster based system compared to the radial [11] G. M. Masters, “Renewable and efficient electric power systems,” 1st ed,
system due to higher cable costs and trenching costs, when Wiley-Interscience, 2004, pp. 307-383.
spread out over the lifetime of the turbines, the annual costs do [12] E. Muljadi, C. P. Butterfield, A. Ellis, J. Mechenbier, J. Hochheimer, R.
Young, N. Miller, R Delmerico, R. Zavadil, and J. C. Smith,
not vary significantly. This results in slightly lower cost of “Equivalencing the collector system of a large wind power plant,” Proc.
generated energy in $/KWh for the proposed cluster based 2006 IEEE Power and Energy Society General Meeting, Montreal,
system compared to the radial configuration. However, with Canada, June 2006.
lower losses, there is greater amount of energy available for
transmission to the grid and therefore higher profits for the
wind farm.

VIII. CONCLUSIONS
This work presented a cluster algorithm based cable layout
design for a large scale wind farm. The proposed design
method yielded lower collector system real power losses
compared to a conventional radial or daisy chain cable layout
method. Although the capital costs with the proposed system
are much higher than conventional radial layout system, the
costs of energy generation in $/KWh from the wind farm do
not vary significantly. In the event of cable failure, in a daisy
chain layout all turbine units beyond the faulted cable go
offline, however, in the proposed pure cluster based method,