Optimal Construction Sequencing for Secant Pile Wall

Jieh-Haur Chen1, Li-Ren Yang2, Mu-Chun Su3, and Jia-Zheng Lin1

1
Institute of Construction Engineering & Management, National Central University, Taoyuan, Taiwan.
2
Department of Business Administration, Tamkang University, Taipei, Taiwan.
3
Department of Computer Science & Information Engineering, National Central University, Taoyuan, Taiwan.
([email protected], [email protected] )

Abstract— Finding the optimal sequence so as to minimizing Comparing to the other two, the evolutionary strategies use
the construction time is one of the solutions to lower the a different view of choosing operators. The particle swarm
construction costs. This study proposes an effective and optimization (PSO) algorithm introduced in 1995 is similar
efficient optimization algorithm, Self-Organizing feather Map
to evolutionary algorithms that initializes with a population
based optimization (SOMO), to minimize the construction time
with an application to a case study in obtaining the optimal of random solutions [5], [6], [7], but is not a simulation by
sequences for both primary and secondary bored piles of a the natural selection. Nevertheless, there are numerous
secant pile wall. The SOMO is a new developed algorithm modifications and applications recently in several research
according to the human brain that is capable of producing fields [8], [9], [10], [11]. Many studies successfully present
topologically ordered mapping, and that can occur in the that simulations and modeling based on self-organizing
evolution of the feature map for optimization. The results
feature map (SOM) are capable of processing applications
demonstrate that the optimal sequences for both primary and
secondary bored piles are determined with 27.21% of time such as vector quantization, adaptive equalization, and
saving. The practicability of the SOMO algorithm is cluster analysis [12]. SOM related studies demonstrate
substantiated. feasibility to solve the optimization problem in terms of the
traveling-salesman problem (TSP) [13], [14]. Recently Su
Keywords - sequencing, optimization, SOM, SOMO has proposed a new optimization algorithm named
algorithm. SOM-based Optimization (SOMO) algorithm to deal with
continuous optimization problems [15], [16].
The purpose of this paper is to apply the SOMO
I. INTRODUCTION algorithm to the construction sequencing optimization with
Reducing pollution, improving waterproofing efficiency, an application to a case study of a secant pile wall. The
and having long durability are three of the keys to construct following session reviews the secant pile wall in connection
retaining walls. Most relatively simple and inexpensive with the construction technique, and then, the SOM
retaining walls still remain the possible problems of air and algorithm is also reviewed. The third session presents the
noise pollution, damage to neighbors, and relatively low development of the SOMO, followed by the application in
quality while performing the work. For particular facilities, the fourth session. The results and discussion are made
which must meet higher requirements of environmental before the last session of the conclusions and suggestions.
protection standards, such as power plants and sewerage
treatment plants, the typical retaining walls may run into II. CONSTRUCTION TECHNIQUE OF SECANT PILE WALL
trouble. Therefore, the secant pile wall is usually a proper The secant pile wall is one type of retaining walls which
answer for those projects requiring higher standards for can provide support for vertical or near-vertical grade
environmental protection. On the other hand, equipment change and prevent from water linking effectively. Such
required to construct the wall is costly and is usually for wall is usually used in heavy construction and infrastructure
rental basis. Time consumption by equipment use makes up projects. Studies regarding the secant pile discuss its
one of major cost drivers. It is always desired to optimize concrete strength during boring and cutting [17], the
the construction time of secant pile walls. problem of the borehole enlargement when applying to
Previous studies have developed optimization algorithms alluvium [18], the feasibility of lateral support for a tunnel
for solving optimization problems. For solving continuous crossing in difficult geologic conditions [19], and a stiff
and discrete problems, the most popular optimization support in to soft to medium stiff clays [20]. Nevertheless,
algorithms are genetic algorithms (GAs) [1], [2], the construction technique of most secant pile walls for
evolutionary programming [3], and evolutionary strategies above-mentioned uses remains similar.
[4]. GAs utilize operators, reproduction crossover, and The typical construction technique involves 7 steps for a
mutation to yield the optimal solutions; yet, the evolutionary unit, which consists of two primary bored piles and one
programming does not equip crosfsover function. secondary bored pile [21]. The first step is to construct a

978-1-4244-2630-0/08/$25.00 ©2008 IEEE 2142

 Proceedings of the 2008 IEEE IEEM

guide wall so as to position the scant pile wall. The next step winner neuron j * at time k, and M × N solutions is used to
is the installation of casing for the first primary bored pile. explore the optimization space. The final iterating step is to
This requires the casing being driven into the ground, return to Step 2 until pre-specified iterations or the
leaving about one meter length of the casing protruding termination criteria are reached.
from the ground. In the third step, the soil within the casing Based on the learning algorithm for forming a SOM and
is removed to form a borehole. It is possible that the casing the SOM original features, conducting the modifications to
is not long enough for designed or required depth in the the SOM is able to solve optimization problems. This new
ground. Bentonite slurry is usually utilized as the support optimization algorithm is named as SOMO algorithm and is
material below the casing. Subsequently, pouring concrete introduced as follows. There are three significant
into the borehole to accomplish the primary bored pile is the modifications made on SOM: the training procedure, lattice
step 4. The second primary bored pile can be completed by initialization method, and projected fitness landscape [15],
repeating Steps 2 to 4. Steps 5 to 7 are the construction of [16].
the secondary bored pile. The casings of the two primary
bored piles need to be pulled out after 24-hour curing time,
at least, is reached. To form a secondary borehole, an auger A. Training Procedure
cuts and removes the soil in between the two primary bored The random initialization method is employed to initialize
piles. The last 2 steps are to place the steel cage into the the weight vectors. Assuming that the optimization problem
secondary borehole and then, pour concrete into the is to maximize the objective function and input
vector I = ( I 1 , " , I n ) = (1 , " ,1 ) , the winning neuron
T T
borehole, which forms the secondary bored pile. Repeating
the steps 2 to 7, the secant pile wall can be constructed.
j* = Argmaxf (wj1 × I1,", wjn × In ) = Argmaxf (wj1 ×1,", wjn ×1)
1≤ j≤M×N 1≤ j≤M×N
III. SOM-BASED OPTIMIZATION ALGORITHM (3)
= Argmaxf (wj1,", wjn) = Argmaxf (wj )
1≤ j≤M×N 1≤ j≤M×N
The principal objective of self-organizing feature maps is
to transform patterns of arbitrary dimensionality into the
responses of one- or two-dimensional arrays of neurons, and is the maximum output criterion where f (w) is the
to perform this transform adaptively in a topological ordered objective function of the optimization problem. Before
fashion [22]. Kohonem in 1995 proposed a learning adjusting the weights of the winner and its neighbors, let
algorithm for forming a feature map. There are 4 steps
including the initialization, winner finding, weight updating, wj(k+1)=wj(k)+λ1Λj*, j[wj* (k)−wj(k)]+λ2(1−Λj*, j)n for1≤ j ≤M×N (4)
and iterating. For the initialization step, randomly choose
values for the initial weight vectors w j (0) . In the next step, where λ1 and λ2 are the learning rates and are usually in the
*
the winning neuron j at time k is found by using either the
ranges of 0 < λ1 ≤ 0.3 and 0 < λ2 ≤ 0.2 , n = (n1 ,", nn ) T is
maximum output criterion or the minimum-distance
Euclidean criterion defined as a noise vector for the new weight vector, and

j* = Arg max ϕ j ( x(k ), w j (k )) = Arg min x(k ) − w j (k ) d j* , j d j* , j (5)

(1) Λ j*, j = 1 − =1−
1≤ j ≤ M × N 1≤ j ≤ M × N
dmax max{d j * ,1, d j * , N , d j * , M ×( N −1) +1 , d j * , M × N }

where x(k ) represents an input pattern; Again, iterating the entire training procedure is required
x(k ) = [x1 (k ),..., x n (k )] is the kth input pattern; ϕ j (⋅, ⋅)
T
until pre-specified iterations or the termination criteria are
reached.
stands for the activation function of neuron j, and ⋅ is the
Euclidean norm. In Step 3, we adjust the weights of the B. Lattice Initialization Method
winner and its neighbors Considering the efficient initialization scheme in a previous
study [23], let the smallest hyper-rectangle of the parameter
wj (k+1) =wj (k)+η(k)Λj*, j (k)[x(k)−wj (k)] for1≤ j ≤M×N (2) space be denoted as [l1 , h1 ]×"× [l n , hn ] where l i is the lowest
valid value of the ith parameter and hi is the highest one.
where η (k ) is usually a positive constant, Λ j* , j ( k ) Supposing that the neurons on the four corners have their
represents the topological neighborhood function of the corresponding weight vectors as follows

2143
 Proceedings of the 2008 IEEE IEEM

w1,1 = (l1 , l 2 , ! , l n ) T (6) proposed lattice initialization method is repeated to

w M , N = (h1 , h2 , ! , hn ) T
(7) re-initialize the M × N neuron lattice. Thus, the local
optimization can be obtained and reserved to the comparison
w1, N = (l1 , l 2 ,! , l n , h n +1 ,! , hn ) T (8) with the others derived from the repeats of the training
¬ 2¼ ¬ 2¼
procedure and lattice initialization method. The optimization
w M ,1 = ( h1 , h2 , ! , h n , l n +1 , ! , l n ) T (9)
¬ 2¼ ¬ 2¼ solution is yielded after the repeats stop founded on the
termination criterion.
We can find the initialization for the weight vectors on the 4
edges, as shown by
IV. APPLICATION
w1, N − w1,1 j −1 N−j
w1, j = ( j − 1) + w1,1 = w1, N + w1,1 , (10) A. Case Introduction and Data Features
N −1 N −1 N −1
for j = 2," , N − 1 Data collected in this research is regarding a construction
project, which includes the secant pile wall of a siphon well
w M , N − w M ,1
for a power plant near the coastline. The siphon well is a
j −1 N−j ,
wM , j = ( j − 1) + w M ,1 = wM ,N + w M ,1 (11) rectangle-shape site with the area of 1,312 square meters. To
N −1 N −1 N −1
for j = 2," , N − 1 construct the secant pile wall, the siphon well requires 104
primary and 103 secondary bored piles. Each pile has 1.2
w M ,1 − w1,1 i −1 M −i meters in diameter and 16 meters in length. There are 3 sets
w i ,1 = (i − 1) + w1,1 = w M ,1 + w1,1 , (12) of equipment performing the work and starting from three of
M −1 M −1 M −1
for j = 2, " , M − 1 the four corners of the construction site. Data for each
primary bored pile have detailed construction time
w M , N − w1, N i −1 M −i , information (in minutes), including 15 activities of grading,
wi,N = (i − 1) + w1, N = w M ,N + w1, N (13) measuring, positioning equipment, driving the first 8-meter
M −1 M −1 M −1
for j = 2,", M − 1 casing into the ground, boring and cutting (0-6 meters),
boring and cutting (6-12 meters), driving the second 8-meter
Thus, we obtain the initialization for the rest neurons using casing into the ground, boring and cutting (12-17 meters),
the following equation measuring the center of the pile, measuring the altitude,
measuring by ultrasound equipment, removing bottom soil,
wi , N − wi ,1 (14)
applying Bentonite slurry, pouring concrete, and removing
wi , j = ( j − 1) + wi ,1 the casing. Similar to the primary bored pile, detailed
N −1
information of construction time in minutes for each
where the initialize process is on the basis of “from top to secondary bored pile is collected and includes the
bottom and from left to right”. We can keep the weight above-mentioned 15 activities plus the activity of placing
vectors from being linearly dependent by randomly adding the steel cage. Thus, for each secondary bored pile, there are
16 activities. In addition, each concrete coagulation time of
noise vector . Accordingly, we define r = (r1 ,", rn ) which
r T
the primary bored pile is obtained.
hi − li There are limitations and assumptions for the collected
−
is a uniformly distributed number between M ×N data. All information on the subject of construction time is
hi − li relating to the wall construction only. Construction time
and M × N . other than that of performing these 16 activities is not
included. For example, time for equipment setting,
C. Projected Fitness Landscape
adjusting, relocating, and idling is not in the scope of data.
The fitness landscape is used to provide a possibly better The congestion that may be caused due to the site layout and
solution by the search of the peak of a specific region. This equipment occupancy is ignored and does not appear in the
broadly searching, so called projected fitness landscape, can data. There is also no sequencing between any primary
be applied to one or multi-dimensional cases, which is an bored pile and secondary bored pile. Once the concrete of
important feature of the SOMO. For the visualization of any two primary bored piles coagulates, the work of the
selecting a specific region, given a small group of neurons in secondary bored pile between these two primary bored piles
the space of M × N , we firstly narrow down a searching can be carried out. The original completion time of the
×Nf
space for the winning neuron and its neighbors of M f . secant wall is 528.85 hours based on 3 sets of equipment
[l f 1 , h f 1 ] ×" × [l fn , h fn ] that perform the activities at the three corners of the
Using the hyper-rectangle of , the

2144
 Proceedings of the 2008 IEEE IEEM

rectangle-shape site, that move clockwise to avoid collision, Table 1

and that construct the secondary bored piles after all primary NEURON WEIGHTS IN SOMO
bored piles are completed. This indicates that the equipment
set 1 constructs the primary bored piles in order of No. 1, 2, Iteration number: 100
3,…, 34 and the equipment set 2 performs from No. 35 to The winner: (9, 9)
No. 70, and so forth. Subsequently, No. 1 to 34, No. 35 to Neuron Neuron Neuron
Weight Weight Weight
69, and No. 70 to 103 secondary bored piles are built in No. No. No.
numeral order by equipment set 1, 2, and 3, respectively. 1 -0.77981418 36 -0.77980232 71 0.45572379
2 -0.77982157 37 -0.77974868 72 0.45577392
3 -0.77983129 38 -0.77980125 73 0.45565149
B. SOMO for minimizing construction time of conducting
4 -0.77982503 39 -0.77983999 74 0.45575526
secant pile wall
5 -0.77976149 40 -0.77980906 75 0.45573786
Following the 4 steps described in the previous sections
6 -0.77976304 41 -0.77974659 76 0.45576620
can compute the optimal construction time using the SOMO
7 -0.77980602 42 -0.77975345 77 0.45572239
algorithm. There are only two limitations for this
8 -0.77979696 43 -0.77983230 78 0.45572990
computation which tries to simulate the reality. One is the 9 -0.77978945 44 -0.77977484 79 0.45568514
number of the equipment sets used in the construction site. 10 -0.77972007 45 -0.77977806 80 0.45567158
Due to the site area, the maximum capacity allowing of 11 -0.77979934 46 -0.77978474 81 0.45570767
simultaneously construction work is 4 sets of equipment but 12 -0.77979416 47 -0.77980471 82 0.45573291
3 sets were utilized throughout the wall construction. The 13 -0.77975708 48 -0.77975798 83 0.45576185
other one is the assumption of sequence. There is no 14 -0.77975053 49 -0.77978897 84 0.45576423
construction sequence for both bored piles except only one 15 -0.77978843 50 -0.77984506 85 0.45568705
condition: Any secondary bored pile can not be built until its 16 -0.77976567 51 -0.77976465 86 0.45576099
two neighbor primary bored piles coagulate. In addition, the 17 -0.77976489 52 -0.77978122 87 0.45573187
pre-specified iteration number is set to 100 with a randomly 18 -0.77975565 53 0.45575458 88 0.45567605
assigned hyper-rectangle of the parameter space where there 19 -0.77980256 54 0.45579866 89 0.45572537
are 104 neurons in the space. The results shown in Table 1 20 -0.77974319 55 0.45579502 90 0.45567277
are the corresponding weights and the global winner neuron 21 -0.77981412 56 0.45574382 91 0.45572296
is at (9, 9). Table 2 demonstrates the construction time and 22 -0.77972031 57 0.45572054 92 0.45571643
the sequences for both primary and secondary bored piles. 23 -0.77972662 58 0.45577070 93 0.45567900
24 -0.77975321 59 0.45567045 94 0.45573813
25 -0.77977014 60 0.45567983 95 0.45575765
V. RESULTS AND DISCUSSION 26 -0.77979714 61 0.45572320 96 0.45567214
The results to evaluate if the SOMO algorithm has 27 -0.77980149 62 0.45578474 97 0.45573005
applicability in solving optimization problems in the field of 28 -0.77981573 63 0.45574453 98 0.45574939
civil engineering can be examined in Table 2. Comparing 29 -0.77979255 64 0.45572564 99 0.45575607
the original construction time with the one shown in Table 30 -0.77979612 65 0.45576674 100 0.45572004
2, 143.92 hours or 27.21% of the original construction time 31 -0.77983040 66 0.45578223 101 0.45576099
can be saved by following the sequences. The computing 32 -0.77973753 67 0.45572257 102 0.45577708
time is less than one second using a personal computer 33 -0.77974707 68 0.45574945 103 0.45575380
equipped with 2.0 Giga CPU and 1 Mega RAM. The 34 -0.77976125 69 0.45568785 104 0.45572272
smallest hyper-rectangle set to 10 by 10 as default with 35 -0.77979648 70 0.45573565
randomly assigned 104 neurons has capable of rapidly
locating the optimization. The sequences for both primary algorithm is applicable.
and secondary bored piles are based on the sorting of the The results reveal that using the SOMO algorithm, which
weights of the corresponding 104 neurons. For example, No. is a new algorithm, has potential in the civil engineering and
50 primary bored pile has the smallest weight, indicating construction areas. There are many issues relating to
that it is the first primary bored pile which should be optimization, to which the SOMO algorithm can apply with
constructed among all 104 piles. Such sorting can easily a few assumptions or restrictions, such as costs, route, and
yield the results shown in Table 2 as the best sequence for resource allocation. However, the application used in this
constructing the secant pile wall. Effectively and efficiently study has assumptions, which lead to further discussions as
solving a time optimization problem by the SOMO follows. The first one is the assumption regarding the

2145
 Proceedings of the 2008 IEEE IEEM

Table 2.
CONSTRUCTION TIME AND SEQUENCE OF PRIMARY AND SECONDARY PILES

Construction time: 384.9333 hours

Computing time: < 1 second(s)
Pile type Sequence (from left to right)
Number of primary bored piles 50 39 43 3 31 4 2 28 1 21 40 7 47 19 36
27 38 11 26 8 35 30 12 29 9 49 15 46 52 45
44 25 16 17 51 6 5 34 48 13 18 42 24 14 37
33 41 20 32 23 22 10 73 59 80 96 90 88 93 60
79 85 69 81 92 100 57 77 67 104 91 61 71 89 64
78 97 87 82 70 75 94 56 63 98 68 103 53 74 99
95 86 101 83 84 76 65 58 72 102 66 62 55 54
Number of secondary bored piles 32 51 48 44 45 33 12 24 27 28 29 8 15 18 9
25 26 49 50 40 46 47 19 88 13 14 78 79 60 87
98 2 3 80 81 23 30 31 103 54 55 37 6 7 10
11 43 16 17 58 59 36 82 20 34 35 102 21 22 83
99 38 39 52 53 67 41 42 1 65 66 4 5 84 85
86 68 100 101 63 64 75 56 57 76 77 94 95 89 72
90 96 97 91 69 61 62 92 93 73 74 70 71

construction time. Only 16 activities plus concrete close to the reality, there above-mentioned impact factors
coagulation are considered in the time computing. Other should be considered; however, some of them are not easy
activities such as equipment movements, redeploying, and to measure and obtain.
interaction by other facilities can easily cause extra time
when the number of the equipment sets exceeds in one.
Although such activities do not appear in the collected data, VI. CONCLUSION
they may affect the construction time of those 16 activities This paper has examined a novel algorithm, SOMO, used
and the degree of their impact remains unknown. Another for solving optimization problems in construction time of
important assumption made in this case study is under the conducting a secant pile wall. The construction technique
given coagulation time of concrete which was measured and and sequence of the secant pile wall are presented in the first
obtained after concrete coagulation had taken place. In fact, place, followed by the introduction of the original SOM
the coagulation time of concrete depends on many concepts. The learning procedures carry out to obtain the
conditions such as humidity, temperature, and water-cement winning neuron so as to output the desired value as well as
ratio. It is difficult to have coagulation time in advance so the optimal solution. In the meanwhile, the iteration concept
that predicting the optimal sequence may be inaccurate.In and fitness landscape are used to seek a possibly better
addition, during the SOMO computation, it is found that the solution around the peak of a specific region. Using
concrete coagulation time may be one of critical factors that collected construction time data of the secant pile wall to
determine the optimal sequences of both primary and second evaluate the SOMO applicability, the results show that
bored piles. Thirdly, the soil condition is not considered. 143.92 hours or 27.21% of the original construction time
The soil under the construction site may have ingredients can be saved, and that the sequences for both primary and
that affect construction time. For example, in the original secondary bored piles can be determined. This substantiates
sequence, the construction time of No. 72 primary bored pile that the SOMO algorithm is capable of dealing with
is about a half hour longer than No. 71’s and No. 73’s. In optimization problems in the construction and civil
the optimal sequence yielded by the SOMO algorithm, No. engineering fields.
71 and 73 primary bored piles are completed ahead of Saving construction time indicates lowering construction
No.72. This may change the soil condition and reduce the costs for most cases but is not always true under the
construction time of No. 72 primary bored pile. The last one condition that does not include all cost drivers. The
is associated with the skill level and physical condition of equipment supplier charges for the total using time. As
the equipment operators. Operators with different level of concerning construction costs, the optimal sequences for
operating skills or under different level of physical both bored piles are unnecessarily the most economical
conditions contribute to the differentiation of the since every second of using equipment counts. Therefore,
construction time. The scope of the application does not considering not only the construction time but all other
include any human effects. To achieve a better outcome

2146
 Proceedings of the 2008 IEEE IEEM

activities that affect the total costs can be closer to the [16] M. C. Su, Y. X. Zhao, and J. Lee, “SOM-based Optimization.” in
2004 IEEE International Joint Conf. on Neural Networks IJCNN,
reality so that the pile sequences found in this study may be
Budapest, Hungary, pp. 781-786, 2004.
changed. One of possible future studies can be carried out to [17] T. Suckling, “Conflicting requirements for 'firm' pile concrete in
improve the practicability by quantifying and adopting all secant pile walls.” Concrete, Vol. 39, No. 6, pp. 26-27, 2005.
other cost drivers. Moreover, the optimal time yielded from [18] S. H. Zhou, J. S. Zheng, Z. G. He, C. G. She, and X. F. Bao, “The
construction method and application of new-style secant piles in
the SOMO algorithm is compared with the original
alluvium.” Zhongguo Tiedao Kexue/China Railway Science, Vol. 27,
construction time. There are other algorithms and methods No. 4, pp. 57-61, 2006.
used to solve optimization problems such as artificial neural [19] T. C. Anderson, and J. L. Williams, “Secant piles support access
networks, case based reasoning, genetic algorithm, ant shafts for tunnel crossing in difficult geologic conditions.”
Proceedings of Sessions of the GeoSupport Conference: Innovation
colony optimization algorithm, particle swarm optimization
and Cooperation in Geo, pp. 299-308, 2004.
algorithm, and so on. Method comparisons based on [20] R. J. Finno, S. Bryson, and M. Calvello, “Performance of a stiff
construction time and cost efficiency can be conducted in support system in soft clay.” Journal of Geotechnical and
the future work. Geoenvironmental Engineering, pp. 128, No. 8, pp. 660-671, 2002.
[21] Land Transport Authority, 2007. http://www.lta.gov.sg/, Singapore.
[22] S. Haykin, Neural Networks, 2nd Ed., Prentice Hall, Upper Saddle
River, NJ, 1999.
REFERENCES [23] M. C. Su, T. K. Liu, and H. T. Chang, “An efficient initialization
[1] J. H. Holland,. Adaptation in Natural and Artificial Systems. scheme for the self-organizing feature map algorithm.” in IEEE Int.
University of Michigan Press, Ann Arbor, Michigan, 1975. Joint Conference on Neural Networks, pp. 1906-1910, 1999.
[2] D. E. Goldberg, Genetic Algorithms in Search, Optimization, and
Machine Learning. Addison-Wesley Publishing Company, Reading,
MA, 1989.
[3] L. J. Fogel, Evolutionary Programming in Perspective: the Top-down
View in Computational Intelligence: Imitating Life. J. M. Zurada, R.
J. Marks II, and C. J. Robinson, Eds., IEEE Press, Piscataway, NJ,
1994, pp. 135-146.
[4] I. Rechenberg, “Evolution Strategy”, in Computational Intelligence:
Imitating Life. in J. M. Zurada, R. J. Marks II, and C. Robinson, Eds.,
IEEE Press, Piscataway, NJ, 1994, pp. 147-159.
[5] J. Kennedy, and R. Eberhart, “Particle swarm optimization,” in IEEE
International Conference on Neural Networks, vol. 4, pp. 1942-1948,
1995.
[6] R. Eberhart, and J. Kennedy, “A new optimizer using particle swarm
theory,” in Proceedings of the Sixth International Symposium on
Micro Machine and Human Science, pp. 39-43, 1995.
[7] J. Kennedy, R. C. Eberhart, and Y. Shi, Swarm Intelligence.
Academic Press, New York, 2001.
[8] X. H. Yuan, C. Wang, Y. C. Zhang, and Y. B. Yang,. “A survey on
application of particle swarm optimization to electric power systems,”
Power System Technology, vol. 28, No. 19, pp. 14-19, 2004.
[9] S. K Goudos, and J. N. Sahalos, “Microwave absorber optimal design
using multi-objective particle swarm optimization.” Microwave and
Optical Technology Letters, Vol. 48, No. 8, pp. 1553-1558, 2006.
[10] H. Zhang, C. M. Tam, H. Li, and J. J. Shi, “Particle swarm
optimization-supported simulation for construction operations.”
Journal of Construction Engineering and Management, Vol. 132,
No.12, pp. 1267-1274, 2006.
[11] C. F. Juang, I. F. Chung, and C. H. Hsu, “Automatic construction of
feedforward/recurrent fuzzy systems by clustering-aided simplex
particle swarm optimization.” Fuzzy Sets and Systems, Vol. 158, No.
18, pp. 1979-1996, 2007.
[12] T. Kohonen,. Self-Organization Maps, Springer-Verlag, Berlin,
Germany, 1995.
[13] B. G. Angeniol, de La C. Vaubois, and J. Y. Le Texier,
“Self-Origanizing Feature Maps and the Travelling Salesman
Problem.” Neural Networks, Vol. 1, pp. 289-293, 1988.
[14] H. D. Jin, K. S. Leung, M. L. Wong, and Z. B. Xu, “An efficient
self-organizing map designed by genetic algorithms for the traveling
salesman problem.” IEEE Trans. on Systems, Man and Cybernetics,
Vol. 33, pp. 877-888, 2003.
[15] M. C. Su, and H. C. Chang, “Fast self-organizing feature map
algorithm,” IEEE Trans. on Neural Networks, Vol. 13, No.3, pp.
721-733, 2000.

2147