You are on page 1of 8

Geotechnique Today- Prediction, Modelling & Construction: Proceedings of Indian Geotechnical Society, Kolkata Chapter, Feb.

28th -March 1st, 2014.

ARTIFICIAL BEE COLONY (ABC) ALGORITHM APPLIED TO SLOPESTABILITY IN SEARCHING THE CRITICAL SURFACE
Dr. Abhijit Saha
Deputy Director (Hydrology), Central Design Office, Irrigation & Waterways Department, Govt. of West Bengal, Kolkata-700091
Email: ovijeet@rediffmail.com

ABSTRACT: The popularity of nature-inspired meta-heuristics in solving complex multidimensional and combinatorial optimization
problems is on the rise. The biologically inspired computation and their solution frameworks abstracted from the natural world are
largely due to its flexibility of hybridization and customization to different problem environments. Swarm intelligence is a research
field that models the collective intelligence in swarms of insects or animals. Artificial Bee Colony (ABC) algorithm is one of the most
recently introduced (Karaboga, 2005) swarm-based algorithms that simulates the intelligent foraging behaviour of a honeybee swarm.
A slight modified version of ABC algorithm proposed by the author is initially tested on some hard benchmark test functions and then
applied in searching the critical failure surface in soil slopes where the independent design variables are the abscissa (CX), ordinate
(CY) of slip circle centre, and depth factor (Nd) of the slip surface. The objective function is the factor of safety (FOS) against failure.
The stability analysis is carried out integrating the present tool with Bishop's simplified method (1955) and results compared with
examples from literature to highlight the robustness of the present methodology. Results produced by ABC and some other natureinspired meta-heuristics like Genetic Algorithm (GA), Genetic Algorithm-Simulated Annealing (GA-SA) hybrid, Big Bang-Big
Crunch (BB-BC) method have been compared (Saha 2003, 2008, 2011, 2013) to demonstrate its versatility.
1

INTRODUCTION

Nature is a principal source of inspiration in devising


optimization models for solving high degree of complex
problems. Swarm intelligence within nature is defined as any
attempt to design algorithms or distributed problem-solving
devices inspired by the collective behaviour of social insect
colonies and other animal societies (Bonabeau et al, 1999). It is
a paradigm that considers collective intelligence as a behaviour
that emerges through the interaction and cooperation of large
numbers of homogeneous lesser intelligent agents in the
environment. Examples include schools of fish, flocks of birds
and colonies of ants. Two fundamental concepts, selforganization and division of labour, are necessary and sufficient
properties to obtain swarm intelligent behaviour such as
distributed problem solving systems that self-organize and adapt
to the given environment in recent years (Karaboga, 2005). The
information is typically stored in the participating homogeneous
agents, or is stored or communicated in the environment through
the use of pheromones in ants, dancing in bees, proximity in
fish and birds and immune system in cells. Like evolutionary
computation, swarm intelligence algorithms are adaptive
strategies that are typically applied to search and optimization
domains. They are characterized by decentralized control, selforganization and adaptation.
In the present study, the performance of ABC algorithm is
initially tested on some hard benchmark optimization test
functions like the Rosenbrocks valley function (Rosenbrock,
1960) and Rastrigin function (Rastrigin, 1974), that yielded fast
and robust convergence * , and then applied to single objective
optimisation of a three dimensional real-world problem: finding
the critical failure surface in soil slope-stability problem.
2

ARTIFICIAL BEE COLONY (ABC) ALGORITHM

ABC algorithm was first proposed by Dervis Karaboga (2005),


which is based on the intelligent behaviour of honeybee swarms
*

elaborated in APPENDIX

finding nectar and sharing the information of food sources with


each other. Honeybee comb build-up and management is a
classic example of teamwork, experience, co-ordination and
synchronization. ABC algorithm has the advantages of strong
robustness, fast convergence, high flexibility and fewer control
parameters. ABC uses a colony of artificial bees in simulating
the behaviour of real honey bees for solving multidimensional
and multimodal optimisation problems. Bee colonies have been
observed to send bees to collect nectar from flower patches
relative to the amount of food available at each patch. Bees
communicate with each other at the hive via a waggle dance
that informs other bees in the hive as to the direction, distance,
and quality rating or fitness of food sources. In the model, the
colony of artificial bees consists of three groups of bees:
employed bees, onlookers and scouts. First half of the colony
consists of employed artificial bees and the second half includes
the onlookers. The search carried out by the artificial bees can be
summarized as follows:
Employed bees determine a food source within the
neighbourhood of the food source in their memory.
Employed bees share their information with onlookers
within the hive and then the onlookers select one of the
food sources.
Onlookers select a food source within the neighbourhood of
the food sources chosen by them.
An employed bee of which the source has been abandoned
becomes a scout and starts to search a new food source
randomly.
Each cycle of the search consists of three steps: (i) moving
the employed and onlooker bees onto the food sources and (ii)
calculating their nectar amounts and determining the scout bees
and then (iii) moving them randomly onto the possible food
sources.
A food source represents a possible solution to the problem
to be optimized. The nectar amount of a food source
corresponds to the quality of the solution represented by that
food source. Onlookers are placed on the foods by using fitnessproportional selection. The mechanism that allows this selection
is similar to a Roulette Wheel that is partitioned into slices.
Each individual has a share directly proportional to its fitness.

69

Geotechnique Today- Prediction, Modelling & Construction: Proceedings of Indian Geotechnical Society, Kolkata Chapter, Feb.28th -March 1st, 2014.

When the roulette wheel is rotated, an individual has a chance of


being selected corresponding to its share of fitness. The selection
probability of each food source is proportional to its fitness as a
ratio to the average fitness of all the flower patches. If the fitness
of a food source is not improved beyond a predetermined limit,
then that food source is abandoned by its employed bee and the
employed bee associated with that food source becomes a scout.
In a robust search process, exploration and exploitation
processes must be carried out together. In the ABC algorithm,
while onlookers and employed bees carry out the exploitation
process in the search space, the scouts control the exploration
process. The classical ABC includes four main phases (Karaboga
& Akay, 2010):
Initialization Phase: The food sources, whose population
size is SN, are randomly generated by scout bees. Each food
source xm is a vector to the optimization problem, xm has D
variables and D is the dimension of searching space of the
objective function to be optimized. The initial food sources are
randomly produced in accordance with:
xm = li + rand(0,1)*(ui li )
(1),
Where, ui and li are the upper and lower bound of the solution
space of objective function, rand(0,1) is a random number within
the range [0,1].
Employed Bee Phase: For each employed bee (whose total
number (BN) is equal to half the number of food sources), a new
source is produced within the neighbourhood of the initial food
source. The higher quantity food source (greedy selection) is
chosen. This food source information stored by employed bee is
shared with the onlooker bees. A neighbouring food source, vmi
is determined and calculated by the following Eqn.(2):

vmi = xmi + mi (xmi -xki )

(2),

Where xk is a randomly selected food source [k {1, 2, . . . ,


BN}], i is a randomly chosen parameter index [i {1, 2, . . .
,D}], mi is a random number within the range [-1,1]. The range
of this parameter can make an appropriate adjustment on specific
issues. The fitness {Fi=fitm(xm)} of food source is essential to
find the global optimal. This fitness is calculated by Eqn.(3).
After that a greedy selection is applied between xm and vm.

1 f x , f m xm 0
m
m
fit m xm

1 f m xm , f m xm 0

(3),

Where, fm(xm) is the objective function value of xm.


The fitness of an individual in slope-stability context in the
present study is calculated in two different ways: in above
traditional way and also employing windowing technique.
In windowing technique, the worst objective function value
of all the food sources/flower patches are subtracted from the
individual food sources objective function value. Hence, the
fitness function {Fi= fitm(xm)} of a food source is: Fi =
max.{fm(x)mi| j = 1, 2, , SN} fm(x)mi , where fm(x)mi is the objective
function of ith food source. The technique is known as
windowing as it eliminates the worst food source-the
probability comes to zero, and stimulates the profitable ones.
Onlooker Bee Phase: The exchange of information among
bees is the most important occurrence in the formation of
collective knowledge. While examining the entire hive, it is
possible to distinguish some parts that commonly exist in all
hives. The most important part of the hive with respect to
exchanging information is the dancing area. Communication
among bees related to the quality of food sources occurs in the
dancing area. Onlooker bees observe the waggle dance
(Waggling is a form of dance in circular direction in the shape

of digit 8. It is repeated again and again by a bee. Its intensity


and direction gives the idea of food source quality and location
to the onlookers) in the dance area of the employed bee and
calculate the profitability of food sources, and then randomly
select a higher food source. Since information about all the
current rich sources is available to an onlooker on the dance
floor, she probably could watch numerous dances and choose to
employ herself at the most profitable source. There is a greater
probability of onlookers choosing more profitable sources since
more information is circulating about the more profitable
sources. Hence, the recruitment is proportional to profitability of
a food source. The quantity of a food source is evaluated by its
own profitability and the profitability of all food sources.
Probability, Pm is determined by the formula:

Pm

fit m xm

fit x

(4),

SN

m 1

Where, fitm(xm) is the fitness of xm. Onlooker bees search the


neighbourhood of food source in accordance with Eqn.(2).
Scout Bee Phase: If the profitability of food source is below
a predetermined "limit" (specified by the user and is dependent
on the complexity of the problem investigated), the solutions will
be abandoned by employed bees. These foragers now become
scout bees and starts randomly searching the neighbourhood for
prospective new food sources/flower patch (new solutions to the
problem investigated). If solution xi has been abandoned, the
new solution xm will be discovered by the scout. The xm is
defined by Eqn.(1).
3

THE MAIN CONCEPTS OF ABC ALGORITHM

Food sources: In nature, the worth of a food source depends


on many aspects, such as its proximity to the nest,
richness/concentration of energy and the ease of energy
extraction. In simple terms, profitability of a food source is
represented by a single quantity. According to different
problems, the initial food sources are randomly generated within
the search space.
Local optimization (search space exploitation) strategy:
Employed foragers are associated with a particular food source,
which they are currently exploiting or are employed at. They
carry the information about this particular source, its distance
and direction from the nest and the profitability of the source and
share this information with a certain probability. In the employed
bee phase, employed bees look for local optimization value in
the neighbourhood of food source. Generally, different local
search strategies are used for different problems.
Random selection strategy in accordance with probability:
Unemployed foragers look for a food source to exploit. There are
two types of unemployed foragers: scouts searching the
surroundings for new food sources and onlookers waiting in the
nest and finding a food source through the information shared by
employed foragers. In the onlooker bee phase, the random
selection strategy is used for local optimization in the
neighbourhood of food source and the higher probability
solution is chosen by onlookers.
Feedback strategy: In scout bee phase, food sources which
are initially poor or have been made poor by exploitation are
abandoned; this means that if a solution cannot be improved
beyond a predetermined limit, new solutions will be
discovered by the scout using the negative feedback strategy
(Karaboga & Akay, 2009).
Global optimization (search space exploration) strategy:
After local optimization and random selection, the global
optimization strategy is used to obtain global optimal value.

70

Geotechnique Today- Prediction, Modelling & Construction: Proceedings of Indian Geotechnical Society, Kolkata Chapter, Feb.28th -March 1st, 2014.

REAL HONEYBEE BEHAVIOUR

The behavioural characteristic of foragers, as described by


Karaboga (2005), is depicted in Fig.-1. Assuming there exist
two discovered food sources as A and B; initially a potential
forager will start as unemployed forager. This bee will have no
knowledge about the food sources around the nest. There are two
possible options for such a bee:
(i) It can be a scout and starts searching around the nest
spontaneously for a food due to some internal motivation or
possible external clue (S in Fig.-1).
(ii) It can be a recruit after watching the waggle dances and
starts searching for a food source (R in Fig.-1).

as 60 m. and 30 m. respectively. To employ the ABC algorithm,


the factor of safety (FOS), F in slope-stability context is defined
in terms of the abscissa (CX) and ordinate (CY) of the circle
centre and the radius of failure surface (R); where, R=f{CX, CY,
NdH}, and Nd is the depth factor. Based on a few trials, the
initial feasible search boundaries of the three independent design
variables, has been identified as: -0.10B CX B, 1.05H CY
3x1.05H, 0.80 Nd 1.25, and depicted in Fig.-2. Till today,
the most extensively used method by the practicing engineer to
find the critical slip surface of a slope is by the Bishop's
simplified method; due to its inherent simplicity in formulation
and easy understanding. The objective function to be minimized,
the FOS (F) is related to the height of slope H, the effective
subsoil parameters c/, / and , the pore pressure ratio ru (= u/h),
the individual slice widths bi, heights hi and i-the slice
inclinations with the horizontal on the failure arc. A total of 15
slices of equal widths are considered in the analysis. Further, the
following are assumed: c//H=0.02, ru=0.5, =40deg. F (=FOS)
is given by Bishop (1955):

c / b i b i h i
sec i
/
1 ru tan

/
H H
i 1 H H
1 tan i tan

.
F
n
b i h i
sin i

i 1 H H
(-0.10B, 3.15H)
INITIAL FEASIBLE SEARCH SPACE
Center
of
circle
CX, CY
i
O

(-0.10B, 1.05H)

0.80 <= Initial Nd<= 1.25

(B, 3.15H)
SUB-SOIL DATA
c/
H = 0.02
/= 40 deg.
u
ub
ru = W = h = 0.50
c/
P
= F + ( l- u)

/
tan
F

(B, 1.05H)
xi

bi
H=30

NH
d

(0,0)

Fig.-1.Behaviour of honeybee foraging for nectar (Karaboga, 2005).

= i i
i
=
i i
i

After finding the food source, the bee utilizes its own capability
to memorize the location and then immediately starts exploiting
it. Hence, the bee will become an employed forager. The
foraging bee takes a load of nectar from the source and returns to
the hive, unloading the nectar to a food store. After unloading
the food, the bee has the following options:
(i) It might become an uncommitted follower after abandoning
the food source (UF).
(ii) It might dance and then recruit nest mates before returning
to the same food source (EF1).
(iii) It might continue to forage at the food source without
recruiting after bees (EF2).
5

SLOPE STABILITY PROBLEM DEFINITION

By the advent of computers, the use of optimization techniques


in locating the critical slip surface has been a major topic for the
researchers. Duncan (1996) presented a comprehensive review
of both limit equilibrium and finite-element analysis of slopes.
Malkawi et al. (2001) developed an effective approach for
locating the critical circular slip surface based on Monte-Carlo
techniques. In this study, a problem cited by Spencer (1967) is
chosen for analysis. The width and height of slope are assumed

i
0.10B=6

B=60

i
i
li

i= i
i i= i

Fig-2. Slope-stability problem definition.


6

IMITATING THE BEE COLONY

Fig.-3 shows the variation of design variables (CX, CY, Nd)


values against minimum objective function f(x)imin., in a ABC
search, captured in 50 successive cycles of a typical simulation
run. The variation is plotted in a concentrically radiating circular
scale, clockwise, with the f(x)imin. corresponding to cycle G1
placed at the apex of the outermost circle. The process clearly
demonstrates the stochastic nature of the algorithm. The global
minimum FOS (Fmin), the corresponding maximum fitness (Fmax)
in windowing technique is shown on top left of the figure. Fig.4 shows the changing value of design variables at local
exploitation of search space with increasing cycles. Fig.-5
portrays the changing maximum fitness (Fmax.) with increasing
cycles. Fmax. adapt to the global optimal with increasing cycles
thrusting the design variables to the best part of the search space
while maintaining its stochastic character. Fig.-6 shows the

71

Geotechnique Today- Prediction, Modelling & Construction: Proceedings of Indian Geotechnical Society, Kolkata Chapter, Feb.28th -March 1st, 2014.

gradual path to optimization of objective function (FOS) value


with increasing cycles.
G1 to G50

1.060444125
1.001917336
1.073072326
1.073072326
1.002094521
1.002486514
1.04667637
10
1.036020143
1.002486514
1.002486514
1.036020143
1.002486514
1.036020143
1.002486514
1.036020143
1.002486514
1.036020143
1
1.002486514
1.037867167
1.003713273
1.037867167
1.008294726
1.037867167
1.008670345
1.037867167
0.1
1.008670345
1.027497541
1.008670345
1.027497541
1.008670345
1.027497541
1.008670345
1.016502211
1.008670345
1.016502211
1.008670345
1.016502211
1.008670345
1.016502211
1.008670345
1.010308419
1.010308419
1.009828788
1.009828788
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419

Fitne ss (F i)=FOSmax.-FOSi(WINDOWING)

48

44

40

36

32

28

Cyc le

24

20

16

0.1

1.08
1.07
1.06
1.05
1.04
1.03
1.02
1.01
1.00
0.99
0.98

G1 to G50

49

45

41

37

33

29

25

21

17

13

Objec tive Func tion, f(x)i= FOSi.

FOS Value

F (i))av.

Initial Mean of Fitness Func tion (Fi) of Food Sources/ Flower Patches
Intermediate
Mean of Fitness Function (Fi) of Food Sources/ Flower Patches
0.50
Final Mean of Fitness Function (Fi) of Flower Patches before Scout Bee Phase

0.48
0.46
0.44
0.42
Random movement with steady
inc rease in average fitness value
(de fine d be low) in suc cessive c yc les
Fitness (F i)=1/ (1+FOSmin.)

0.40
0.38
0.36

1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49

Fig.3. The min. obj. func.[f(x)imin. =FoSmin] vs. a design variables


(CX, CY & Nd).

Obje c tive Func tion=FoS

Fig.6. The gradual path to optimization of objective function


(FOS) with increasing cycles.

Av. Fitness Function (

Min.FoS Vs. Nd

Objective Function=FOSS

Cyc le

1.060444125
1.001917336
1.073072326
1.002094521
1.073072326
1.002486514
1.04667637
1.002486514
1.036020143
1.00
1.002486514
1.036020143
1.002486514
1.036020143
1.002486514
1.036020143
0.95
1.002486514
1.036020143
1.002486514
1.037867167
1.003713273
1.037867167
0.90
1.008294726
1.037867167
1.008670345
1.037867167
0.85
1.008670345
1.027497541
1.008670345
1.027497541
1.008670345
1.027497541
1.008670345
1.016502211
1.008670345
1.016502211
1.008670345
1.016502211
1.008670345
1.016502211
1.008670345
1.010308419
1.009828788
1.010308419
1.009828788
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419

Cyc le

Fig.7. The mean of fitness function (fi av.) at local exploitation of


search space.
Loc al optimization (SearchSpace
Exploitation) at end of eac h c ycle

49

46

43

40

37

34

31

28

25

19

16

13

10

Cyc le

Fig.4. The changing value of design variables at local


exploitation of search space with increasing cycles.

Space,

49

45

41

37

33

29

25

21

17

13

0.1

CX at Local Optimization
CY at Local Optimization
Nd at Local Optimization

Change in Mean Value of


Objective Function (FOS) after
Local Exploitation of Search
Space in sucessive cycles

f(x)

Initial Mean of Food Sources/ Flower Patches (FOS)


Intermediate Mean of Food Sources/ Flower Patches (FOS)
Final Mean of Flower Patches (FOS) before Scout Bee Phase

Mean of Objective Func. (FOS)


afterLocal Exploitation of Search

(Employed Bee &OnlookerBee


phases) &before Sc out Bee
(SearchSpaceExploration)phase

10

22

100

Design Variables (CX, CY, Nd)


before Sc out Bee Phase

Fig.5. The change in max. fitness in windowing technique (Fmax)


with increasing cycles.

Min.FoS Vs. CY
Fmin.=1.002095
F max.=1.02663

10

12

G1 to G50

1.060444125
1.001917336
1.073072326
1.002094521
1.073072326
100
1.002486514
1.04667637
1.002486514
1.036020143
95
1.002486514
1.036020143
1.002486514
1.036020143
90
1.002486514
1.036020143
85
1.002486514
1.036020143
80
1.002486514
1.037867167
75
1.003713273
1.037867167
70
1.008294726
1.037867167
65
60
1.008670345
1.037867167
55
1.008670345
1.027497541
1.008670345
1.027497541
1.008670345
1.027497541
1.008670345
1.016502211
1.008670345
1.016502211
1.008670345
1.016502211
1.008670345
1.016502211
1.008670345
1.010308419
1.009828788
1.010308419
1.009828788
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419

Fmin.=1.002095
F max.=1.02663

Change in maximum fitness


at suc c e ssive c ycles

Min.FoS Vs. CX

100

Max. Fitness of Food


Sources/ FlowerPatc hes after
Loc al Exploitation of Search
Spac e in Eac h Cyc le
F(x)max. =Max.[ f(x)max. -f(x)i]

Max.Fitness of Initial Food Sources/ Flower Patches


Max.Fitness of Intermediate Food Sources found by ABC
Max.Final Fitness of Food Sources found by ABC

1.060444125
1.073072326
1.036020143
1.036020143
1.036020143
1.037867167
1.037867167
1.027497541
1.016502211
1.016502211
1.010308419
1.010308419
1.010308419
1.010308419
1.010308419
1.009828788
1.008670345
1.008670345
1.008670345
1.008670345
1.008294726
1.002486514
1.002486514
1.002486514
1.002486514
1.001917336

Fmin.=1.002095
F max.=1.02663

The steady decrease in objective function value with


increasing cycles signifies the fast movement of variables to the
global optimum. Fig.-7 shows the mean of (traditional) fitness
function (Fi av.) at local exploitation of search space with
increasing cycles. Random movement with steady and gradual
increase of fitness in successive cycles is evident. Fi av. gets
refined with increasing cycles maintaining the algorithms
heuristic character.

Cyc le

Fig.8. The stochastic decrease in av. f(x)i with increasing cycles.

72

Geotechnique Today- Prediction, Modelling & Construction: Proceedings of Indian Geotechnical Society, Kolkata Chapter, Feb.28th -March 1st, 2014.

Searc h Spac e,

f(x)

10

0.1
Change in Standard Deviation of Objective Function (FOS)
after Local Exploitation of Search Space in sucessive cycles

49

46

43

40

37

34

31

28

25

22

19

16

13

10

0.01
Cyc le

Fig.9. The stochastic decrease of std. dev. of obj. func. { f(x)i}


after local exploitation of search space with increasing cycles.

(Fig.b)], with increasing cycles illustrates the generate-and-test


heuristic search technique that exploit domain-specific
knowledge. Fig.12 shows the design tri-variables adapting to
move to the best part of search space. Fig.13 depict search space
exploration by random searching the environment by Scout Bees
with a view to find profitable food sources/flower patches,
whose number decreases with increasing cycles as global
optimal is reached. Fig.14 portrays superimposed results of four
simulation runs as regards the maximum fitness of food sources
after local exploitation of search space. Fig.15 shows the
artificial intelligent character of the algorithm and quick
movement of solutions to best part of the search space with
increasing cycles, keeping provision for further exploration of
search space by the scout bees. Fig.16 shows the comparison of
ABC results with various other methods.
Initial Random Food Source (CX) within the Search Space

11

1.04
1.03

7
6
5
4
3

1.02

1.01

49

47

45

43

41

39

37

35

33

31

29

27

25

21

19

17

15

13

23

49

47

45

43

41

39

37

35

33

31

29

27

25

23

21

19

17

15

The initial, intermediate and final average objective function


(Fav.=f(x)i/SN) of each cycle and the standard deviation of f(x)i
versus the cycle of evolution are shown in Fig.-8 and Fig.-9
respectively. Fig.10 shows the fast convergence to the best food
source by the employed bees and onlooker bees with increasing
cycles. Fig.11 depict the stochastic movement of average fitness,
F(x)i,av. in windowing technique (Fig.a), as well as in traditional
way (Fig.b), with increasing cycles. The stochastic movement of
average fitness and the gradual decrease in windowing
technique (Fig.a), [and gradual increase in traditional way

49

47

45

43

41

39

37

35

33

31

Cyc le

Fig.11. The stochastic movement of average fitness, F(x)i, av.


[(a): gradual decrease in windowing technique, (b): gradual
increase in traditional way] with increasing cycles.

1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49

0.80

29

0.82

Fitness (F i)=1/ (1+FOSmin.)

0.19

27

Greedy Selection by Employed Bees between Initial Random Food Source &
Random Food Source selected in its Neighbourhood
Greedy Selection by Onlooker Bees betn. their Initial Random Food Source &
Source shown by Employed Bees, based on Fitness Probability
Abondoned Food Source (Nd) replaced by Random Food Source within Search
Space for Scout Bees for next phase

0.84

25

New Random Food Sources selected by Employed Bees in the neighbourhood

0.86

23

Initial Random Food Source (Nd) within the Search Space

0.88

21

0.90

19

Random move me nt with steady &


gradual inc rease in fitness value
(de fine d be low) in suc cessive c yc les

0.92

17

0.22

0.94

13

0.23

0.96

11

Design Variable, Nd

Av. Fitness Func tion, (Fi)av.

0.98

0.24

1.00

0.25

0.20

Cyc le

1.02

Cyc le

Av. Fitness of Solutions found by Employed Bee


Av. Fitness of Solutions found by Onlooker Bee

0.21

11

1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49

13

Fitness (F i)=FOSma x.-FOSi(WINDOWING)

11

Greedy Selection by Employed Bees between Initial Random Food Source &
Random Food Source selected in its Neighbourhood
Greedy Selection by Onlooker Bees betn. their Initial Random Food Source &
Source shown by Employed Bees, based on Fitness Probability
Abondoned Food Source (CY) replaced by Random Food Source within Search
Space for Scout Bees for next phase

New Random Food Sources selected by Employed Bees in the neighbourhood

Random movement with steady &


gradual dec rease in fitness value
(in Windowing Te c hnique) in
suc c essive c yc les

99
97
95
93
91
89
87
85
83
81
79
77
75
73
71
69
67
65

Design Variable, CY

Av. Fitness of Solutions found by Employed Bee


Av. Fitness of Solutions found by Onlooker Bee

0.1

Cyc le

Initial Random Food Source (CY) within the Search Space

Fig.10. Fast and defined convergence to the best food source


(global optimization) by the employed bees and onlooker bees.

10

1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49

Cyc le

100

1.00

Av. Fitness Func tion, (Fi)av.

1.05

Best find by Employed Bee


Best find by Onlooker Bee

1.06

1.07

New Random Food Sources selected by Employed Bees in the


neighbourhood
Greedy Selection by Employed Bees between Initial Random Food Source
& Random Food Source selected in its Neighbourhood
Greedy Selection by Onlooker Bees betn. their Initial Random Food
Source & Source shown by Employed Bees, based on Fitness Probability
Abondoned Food Source (CX) replaced by Random Food Source within
Search Space for Scout Bees for next phase

10

Fast movement to best part of the


searc h spac e in suc c essive c ycles by
the Employed &OnlookerBees

Design Variable, CX

Objec tive Func tion (f(x)=FOS)

1.08

15

Standard Deviation of Obj. Func t.


(FOS) afterLoc al Exploitation of

Initial Std. Dev. of Food Sourc es/ Flower Patc hes (FOS)
100
Intermediate Std. Dev. of Food Sourc es/ Flower Patc hes (FOS)
Final Std. Dev. of Flower Patc hes (FOS) before Sc out Bee Phase

Cyc le

Fig.12. Design Tri-variables: (a):- CX, (b):- CY, (c):- Nd;


Adapting to move to the best part of search space.
7

CONCLUSION

ABC algorithm being initially successfully tested on some hard


benchmark test functions , is applied to slope-stability problem
and compared with that of ordinary grid search, variational
method and nature inspired (GA, GA-SA hybrid, BB-BC)
artificial intelligent population-based algorithms in searching the

Detailed in APPENDIX

73

100
90
80
70
60
50
40

Number of New Food Sources explored by Scout Bees


1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49

60

50

40

70
60
50
40

90
80
70
60
50
40

60

50

40

30

20

10

G30:Employed Bees
Design Variable, CX
G30:OnlookerBees
100
90
80
70
60
50
40
60

50

40

30

20

10

60

50

40

30

20

10

30

G40:Employed Bees
Design Variable, CX
G40:OnlookerBees

80

-10

100

30

90

60

50

40

30

20

10

G20:Employed Bees
Design Variable, CX
G20:OnlookerBees

100

30
-10

30

10

30

G10:Employed Bees Design Variable, CX


G10:OnlookerBees
Design Variable, CY

Design Variable, CY

-10

60

50

40

30

20

10

-10

30

20

40

50

10

Design Variable, CY

60

12

70

100
90
80
70
60
50
40
30

-10

14

80

Design Variable, CY

Search Space Exploration

16

90

-10

Number of New Food Sources


found by Scout Bees at end on
each search cycle, (N)

18

100

G1: Employed Bees


Design Variable, CX
G1: OnlookerBees

Design Variable, CY

critical slip surface of a vulnerable soil slope. From the


simulation results it emerged that the proposed algorithm has the
ability to get out of the local minimum and can efficiently be
used for multivariable, multimodal function optimization. The
spectrum of application area of ABC is widespread since it has
the inherent potential of optimizing any multi-dimensional and
multi-modal function. Multivariable functions; both continuous
and discontinuous can be programmed. Function value
evaluations at discrete points only enable it to handle nondifferentiable functions at ease. A typical trial with 50 cycle runs
produced fine convergence. A typical soil-slope investigated has
revealed that minimum FOS to be about 7% less than that of a
directed grid search (Spencer, 1967) and 4-5% less than that of
GA (Saha, 2003, 2008) and hybrid GA-SA (Saha, 2011a) and
slightly less than Big Bang-Big Crunch (Saha, 2011b).
Moreover, it is 13% less than that obtained by variational
method (Narayan et al, 1976).

Design Variable, CY

Geotechnique Today- Prediction, Modelling & Construction: Proceedings of Indian Geotechnical Society, Kolkata Chapter, Feb.28th -March 1st, 2014.

G50:Employed Bees
G50:OnlookerBees Design Variable, CX

Fig.15. CX Vs. CY: The artificial intelligent character


depiction. Movement of solutions to best part of the search space
[a-G1, b-G10, c-G20, d-G30, e-G40, f-G50].

Cycle (G)

Fig.13. Global search: Search Space Exploration by random


searching the environment by Scout Bees with a view to find
profitable food sources/flower patches, whose number
decreases with increasing cycles.

Spenc er(1967) :1.07


Narayan et.al. (1976) :1.13
Saha (2011): GA (3 Variables) (n=20, Gen.=75) :1.0104
Saha (2011): GA-SA Hybrid (3 Variables) (n=20, Gen.=75) :1.00249
Saha (2012):BB-BC (3 Variables) (n=30, Gen.=35) :1.0045

- 4.6628, 94.3152
90

1.9062, 77.9340
1.5817, 78.7950
3.1973, 75.0109

80

10

Present Study:ABC (3 Variables) (n=30, Gen.=35) :1.002095

Slope Profile
SpencerArc
SpencerCenter
GA Arc (n=20, Generation=70th)
GA Center(n=20, Generation=70th)
GA-SA Hybrid: Arc (n=20, Generation=37th)
GA-SA Hybrid: Centre (n=20, Generation=37th)
BB-BC(ES) Arc (n=30, Generation=43)
BB-BC(ES) Center(n=30, Generation=43)
ABC Center(n=20, Generation=50)
ABC Arc (n=20, Generation=50)

70

Height in metres

Max. Fitness of Food Sources/Flower


Patches after Local Exploitation of
Search Space in Each Cycle (F max. )

Minimum Factorof Safety (FOSmin)

100

9.6317, 63.7615
60

ARTIFICIALBEE COLONY

50

CX=1.582, CY=78.795, Nd=0.9939


R=f(CX,CY,Nd) =94.3042
40

FoS=1.002095

0.1

Fitness (F i)=FOSmax.-FOSi(WINDOWING)
1st Simulation Run
3rd Simulation Run

20

2nd Simulation Run


4th Simulation Run

Genetic Alg. (Roulette Wheel Selection)

SOILDATA
B = 60 m
H = 30 m
= 26.50 deg.
c//H = 0.02

30

CX=1.906, CY=77.934, Nd=0.965


R=f(CX,CY,Nd) =76.87

FoS=1.0104
Genetic Algorithm-Simulated Annealing Hybrid
CX=- 4.663, CY=94.315, Nd=0.995
R=f(CX,CY,Nd) =94.304

10

FoS=1.00249

1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49

0.01

Cyc le

0
-5
-10

Fig.14. Superimposed results of four simulation runs.

15

25

35

45

55

65

75

BIG BANG- BIG CRUNCH (with Elitist Strategy)


CX=3.1973, CY=75.0109, Nd=0.9980
85
95
R=f(CX,CY,Nd) =75.0147

FoS=1.0045

Distance in metres

Fig.16. Comparison of results.


8

APPENDIX

In mathematical optimization, the Rosenbrock function (Howard


H. Rosenbrock, 1960) is a non-convex function and a classic
optimization problem (also known as the second function of De
Jong). It is also called Rosenbrock's valley or Rosenbrock's
banana function. The global minimum is inside a long,
narrow, parabolic shaped flat valley. The function has the
following definition:

f x

100 x
n 1
i 1

i 1

2
1

1 x i

To find the valley is trivial, however convergence to the


global optimum is difficult and hence this problem is preferred
frequently by researchers to test the performance of optimization
algorithms. The test area is usually restricted to hypercube:
2.048 xi +2.048; where, i=1,2,3,.n.
It has a global minimum of f(x) = 0, obtainable at xi =1
(i=1,2,3,..n).

The Rastrigin function is another non-convex, non-linear


multimodal function. Finding the minimum of this function is a
fairly difficult problem due to its large search space and its large
number of local minima. It was first proposed by Rastrigin
(1974) as a 2-dimensional function and has been generalized by
Mhlenbein et al. (1991). It is defined by:

f x 10n

x
n

i 1

2
i

10Cos 2 x i ,

-5.12 xi +5.12; where, i=1,2,3,.n. This function is


highly multimodal. It has a global minimum of f(x) = 0,
obtainable at xi =0 (i=1,2,3,..n).
The 2-D plots of Rosenbrock function and Rastrigin function
are shown in Fig.17 and Fig.18. In the current study, the 3-D
versions of above functions are explored, as the slope-stability
problem is typified by 3 design variables: CX, CY, Nd. Fig.21 to
Fig.28 unfold the powerful features of the ABC algorithm, as
captured in the path to optimization and fast yet robust
convergence of these two difficult benchmark test functions.

74

100

Rastrigin's Func tion),

f(x)

Mean of Objec tive Func tion (i.e.,

Geotechnique Today- Prediction, Modelling & Construction: Proceedings of Indian Geotechnical Society, Kolkata Chapter, Feb.28th -March 1st, 2014.

10
Change in Mean Value
of Objec tive Func tion in
suc c essive c yc les

1
0.1
0.01

Initial Mean of Obj. Function


Intermediate Mean Of Obj. Function
Mean of Final Obj. Function

0.001

0.0001

0
4
8
12
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80
84
88
92
96
100

Fig.17. Plot of Rosenbrock Function with two variables (the


optimum lie in the curvilinear deep blue narrow valley-line).

Cyc le

(i.e., Rastrigin's Func tion),

f(x)

Standard Deviation of Obj. Func .

Fig.21. Change in Mean of Obj. Func. at successive cycles.


100
10
Change in Standard
Deviation of Objective
Func tion in suc c essive c ycles

1
0.1
0.01
0.001

Initial Std. Dev. of Obj. Function


Intermediate Std. Dev. Of Obj. Function
Std. Dev. Of Final Obj. Function

0
4
8
12
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80
84
88
92
96
100

0.0001

Cyc le

Fig.18. Plot of Rastrigin Function with two variables (the


optimum lie at the centre amongst numerous local optima).

Fig.22. Change in Standard Deviation of Obj. Func. at


successive cycles.

Av. Probability of Fitness


Func tion, pav. [F i]

100
10
1

Best find by Employed Bee


Best find by OnlookerBee

0.1
0.01
0.001

0.00001

0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
160
170
180
190
200

0.000001

Cyc le

Fig.19. Fast and defined convergence to the best food source


(global optimization) by the employed and onlooker bees.
Fitness (F i) = f(x)max.-f(x)i(WINDOWING TECHNIQUE)

Max. Fitness of Food Sources/Flower


Patches at Local Exploitation of
Search Space in Sucessive Cycles
(F max . )

10000

1st Simulation Run


2nd Simulation Run
3rd Simulation Run

1000
100

1.00
0.96
0.92
0.88
0.84
0.80
0.76
0.72
0.68
0.64
0.60
0.56
0.52

Av. Probability of Fitness Function (Fi)

Fitness (F i) = f(x)max.-f(x)i
(WINDOWING TECHNIQUE)
Random c hange in Av. Probability of
fitness func tion in suc cessive c yc les
0
4
8
12
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80
84
88
92
96
100

Gradual movement and


fast c onvergenc e to the
best part of searc h space
in suc c essive c ycles

0.0001

10
1

0.1

Cyc le

Fig.23. Increase in average probability of fitness function with


increasing cycles.
Objective Function ,
f(x)=Rastrigin's Function)

(f(x)=Rosenbroc k's Func tion)

1000

100
10
1
0.1

Gradual movement and


fast c onvergenc e to the
best part of searc h space
in suc c essive c ycles

0.01
0.001
0.0001
0.00001

0.000001

0.001

0.0000001

Best find by Employed Bee


Best find by Onlooker Bee
0
4
8
12
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80
84
88
92
96
100

0.01

200

190

180

170

160

150

140

130

120

110

90

100

80

70

60

50

40

30

20

10

0.0001

Cyc le (G)

Fig.20. The decrease in maximum fitness (Fmax.) (in windowing


technique) of final population with increasing cycles .

Cycle

Fig.24. Fast and defined convergence to the best food source


(global optimization) by the employed bees and onlooker bees.

75

Geotechnique Today- Prediction, Modelling & Construction: Proceedings of Indian Geotechnical Society, Kolkata Chapter, Feb.28th -March 1st, 2014.

Av. Fitness Function, F

i a v.

100

Fitness (F i) = f(x)max.-f(x)i(WINDOWING TECHNIQUE)


Random movement with
steady &gradual
dec rease in av. fitness
value in suc c essive c ycles

10
1
0.1
0.01

Av. Fitness of Solutions found by Employed Bee


Av. Fitness of Solutions found by Onlooker Bee
0
4
8
12
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80
84
88
92
96
100

0.001

Cycle

Search Space Exploration

15
13
11
9

Number of New Food


Sources for Scout Bees

7
5
3
1

0
4
8
12
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80
84
88
92
96
100

Numberof Food Sourc es for


Sc out Bees at end on each
searc h c yc le

Fig.25. Exponential decrease in average fitness with increasing


cycles by the employed bees and onlooker bees, while
maintaining its heuristic character.

Cyc le

Fig.26. Search Space Exploration by random searching the


environment by Scout Bees to find profitable food sources,
whose number decreases with increasing cycles.

Rastrigin's Func tion

Change in value of Objec tive Func tion at suc cessive c ycles


100
10
1
0.1
0.01
0.001
0.0001
0.00001
0.000001

1st Simulation Run


2nd Simulation Run
3rd Simulation Run
0
4
8
12
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80
84
88
92
96
100

0.0000001
Cyc le

Fig.27. Three typical simulation runs showing the random path


to convergence of the Rastrigins function in ABC optimization.
Max. Fitness of Food Sources/Flower
Patches at Local Exploitation of
Search Space in sucessive cycles
(F max . )

100

Fitness (F i) = f(x)max.-f(x)i(WINDOWING TECHNIQUE)

10
1

0.1

0.01

1st Simulation Run


2nd Simulation Run
3rd Simulation Run
0
4
8
12
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80
84
88
92
96
100

0.001

Cyc le (G)

Fig.28. The decrease in maximum fitness (Fmax.) (in windowing


technique) of final population with increasing cycles.
9

REFERENCES

Bishop A.W. (1955). The use of slip circles in the stability


analysis of slopes. Geotechnique. Vol.5, 7-17.

Bonabeau E., Dorigo M., Theraulaz G. (1999). Swarm


Intelligence: From Natural to Artificial Intelligence, NY:
Oxford University Press, New York.
Duncan J. M. (1996). State of the art: Limit equilibrium and
finite element analysis of slopes. Journal of Geotechnical
Engineering, ASCE, 122(7), 577596.
Karaboga Dervis (2005). An idea based on honey bee swarm for
numerical optimization. Technical Report TR06, Erciyes
University, Engineering Faculty, Computer Engg. Dept.
Karaboga Dervis & Basturk Bahriye (2007). A powerful and
efficient algorithm for numerical function optimization:
Artificial bee colony (ABC) algorithm. Journal of Global
Optimization, 39(3), 459471.
Karaboga Dervis & Basturk Bahriye (2007). Advances in Soft
Computing: Foundations of Fuzzy Logic and Soft Computing,
volume 4529/2007 of LNCS, chapter Artificial Bee Colony
Optimization
Algorithm
for
Solving
Constrained
Optimization Problems, 789798. Springer-Verlag.
Karaboga Dervis & Basturk Bahriye (2008). On the performance
of artificial bee colony (ABC) algorithm. Applied Soft
Computing, 8(1), 687697.
Karaboga Dervis and Akay Bahriye (2009). A comparative study
of artificial bee colony algorithm. Applied Mathematics and
Computation, 214, 108-132.
Karaboga Dervis & Akay Bahriye. (2010). Artificial Bee Colony
(ABC), Harmony Search & Bees Algorithm on Numerical
Optimization.
Basturk Bahriye & Karaboga Dervis (2006). An artificial bee
colony (ABC) algorithm for numeric function optimization.
IEEE Swarm Intelligence Symp., Indianapolis, Indiana, USA.
Karaboga Dervis & Basturk Bahriye. (2008). On the
performance of artificial bee colony (ABC) algorithm.
Applied Soft Computing, Vol.-8, 687697.
Malkawi A.H., Hassan W.F., and Sarma S.K. (2001). Global
search method for locating general slip surface using Monte
Carlo techniques. Journal of Geotechnical & Geoenvironmental Engg., ASCE, 127(8), 688-398.
Narayan C.G.P., Bhatkar V.P. & Ramamurthy T. (1976). Slope
stability analysis by variational method. Indian Geotechnical
Journal, Vol. 6, 68-90.
Rosenbrock, H. H. (1960), "An automatic method for finding the
greatest or least value of a function", The Computer Journal,
3: 175184.
Rastrigin, L. A. (1974). "Systems of external control".
Saha Abhijit (2003). Genetic algorithm based search in slope
stability analysis. Proceedings of 12th Asian Regional Conf. in
Soil Mechanics and Geotechnical Engineering, Singapore,
Vol.-I, 981-984.
Saha Abhijit (2008). Genetic algorithm as a function optimizer
in solving the tri-variate slope-stability problem. National
Seminar on Geotechnique in Present Development
Scenario, Kolkata, India.
Saha Abhijit (2011a). Genetic algorithm based search coupled
with Boltzmann selection in locating the critical surface in
slope stability. Proceedings of 14th Asian Regional
Conference on Soil Mech. & Geotechnical Engineering,
Hong Kong, China, Paper No.115, 172-177.
Saha Abhijit (2011b). Adaptive elitist-population based BigBang Big-Crunch optimization algorithm applied to slopestability. Proceedings of Indian Geotechnical Society,
Kolkata Chapter, India, Paper No.7.
Saha Abhijit (2013). Big-Bang Big-Crunch optimization in
locating the critical surface in slope-stability. Proceedings of
Indian Geotechnical Conference-2013, Geotechnical
Advances and Novel Geomechanical Applications, ISBN:
978-81-925548-1-5, Paper No.-82, Roorkee, India.
Spencer E. (1967). A Method of Analysis of the Stability of
Embankments assuming Parallel Inter-Slice Forces.
Geotechnique, London, Vol.17, 11-26.

76

You might also like