EMODS: A NOVEL EVOLUTIONARY METAHEURISTIC BASED IN
THE AUTOMATA THEORY FOR THE MULTIOBJECTIVE
OPTIMIZATION OF COMBINATORIALS PROBLEMS
Elias David Nino Ruiz
1,2
and Anangelica Isabel Chinchilla Camargo
2
1
Virginia Polytechnic Institute and State University, Blacksburg, Virgina, U.S.A.
2
Universidad del Norte, Barranquilla, Atlantico, Colombia
Keywords:
Combinatorial optimization, Metaheuristic, Evolutionary rules, MultiObjective optimization, Traveling
salesman problem.
Abstract:
This paper states a novel Evolutionary Metaheuristic based in the Automata Theory for the Multiobjective
Optimization of Combinatorial Problems named EMODS. The proposed algorithm uses the natural selection
theory to explore the feasible solutions space of a Combinatorial Problem. Due to this, local optimums are
avoided. Also, EMODS takes advantage in the optimization process from the Metaheuristic of Deterministic
Swapping to avoid finding unfeasible solutions. The proposed algorithm was tested using well known instances
from the TSPLIB with three objectives. Its results were compared against four Multiobjective Simulated
Annealing inspired Algorithms using metrics from the specialized literature. In every case, the EMODS
results on the metrics were always better and in some of those cases, the distance from the Real Solutions was
4%.
1 INTRODUCTION
Combinatorial optimization is a branch of optimiza-
tion. Its domain is optimization problems where the
set of feasible solutions is discrete or can be reduced
to a discrete one, and the goal is to find the best pos-
sible solution (Yong-fa and Ming-yang, 2004). In this
field it is possible to find a lot of problems denomi-
nated NP - Hard such as Multi-depot vehicle routing
problem (Lim and Wang, 2005), delivery and pickup
vehicle routing problem with time windows (Wang
and Lang, 2008), multi-depot vehicle routing problem
with weight-related costs (Fung et al., 2009), Railway
Traveling Salesman Problem (Hu and Raidl, 2008),
Heterogeneous, Multiple Depot, Multiple Traveling
Salesman Problem (Oberlin et al., 2009) and Travel-
ing Salesman with Multi-agent (Wang and Xu, 2009).
One of the most classical problems in combinato-
rial optimization is the Traveling Salesman Problem
and it has been analyzed for years (Sauer and Coelho,
2008) either in a mono or multi - objective way. Al-
though several algorithms have been implemented to
solve TSP, there is no one that optimal solves it in a
polynomial time.
This paper is structured as follows. In Section 2
some fundamentals concepts such as Multiobjective
Optimization and Genetic algorithms are reviewed. In
Section 3, a novel evolutionary metaheuristic is de-
fined on the MODS template. Lastly, in Section 4,
the metaheuristic proposed is tested and its results are
analyzed.
2 PRELIMINARIES
2.1 Multi - objective Optimization
The multi - objective optimization consists in two or
more objectives functions to optimize and a set of
constraints (Glover and Laguna, 1997):
Opt. F(X) = { f
1
(X), f
2
(X), f
3
(X), ..., f
n
(X)} (1)
Subject to
H(X) = 0 (2)
G(X) 0 (3)
X
l
X X
u
(4)
X is the set of desition variables of the problem.
F(X) is the set of objective functions.n is the num-
ber of objective functions. H(X) and G(X) are the
constraints of the problem. Finally, X
l
and X
u
are the
lower and upper bound, respectively, of X.
399
Nino Ruiz E. and Chinchilla Camargo A..
EMODS: A NOVEL EVOLUTIONARY METAHEURISTIC BASED IN THE AUTOMATA THEORY FOR THE MULTIOBJECTIVE OPTIMIZATION OF
COMBINATORIALS PROBLEMS.
DOI: 10.5220/0003754003990404
In Proceedings of the 1st International Conference on Operations Research and Enterprise Systems (ICORES-2012), pages 399-404
ISBN: 978-989-8425-97-3
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
2.2 Pareto Front
As well known, a Pareto Front is a set of nondomi-
nated solutions; it means that all the solutions of the
PF are optimal. In the particular case of the three - ob-
jective optimization of a combinatorial problem, the
PF will be in
3
as can be seen in figure 1. Each
point of this set represents a solution for the problem.
Therefore, the dimension of the Pareto Front depends
on the number of objectives of the problem.
0.4
0.6
0.8
1
1.2
1.4
1.6
x 10
5
0.5
1
1.5
x 10
5
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
x 10
5
f1
f2
f3
No Dominated Solutions
Figure 1: Pareto Front for a particular three - objective prob-
lem.
2.3 Genetic Algorithms
Genetic Algorithms (GA) are algorithms inspired in
the natural selection theory. They consist in three
steps as follows:
Step 1. Selection. Select solutions from a popula-
tion.
Step 2. Crossover. Cross the selected solutions
avoiding local optimums.
Step 3. Mutation. Perturbs the new solutions found
for increasing the population.
The most known Genetic Algorithms from the
literature (Glover and Laguna, 1997) are the Non-
Dominated Sorting Genetic Algorithm (NSGA - II)
and the Strength Pareto Evolutionary Algorithm 2
(SPEA2). NSGA II use a no - dominated sort for
sorting the solutions in different Pareto Sets. Con-
sequently, it demands a lot of time, but it allows a
global verification of the solutions for avoiding the
Local Optimums. On the other hand, SPEA2 is an
improvement of SPEA. The difference with the first
version is that SPEA2 works using strength for ev-
ery solution according to the number of solutions
that it dominates. Consequently, at the end of the
iterations, SPEA2 has the non dominated solutions
stronger avoiding Local Optimums.
2.4 Metaheuristic Of Deterministic
Swapping
Metaheuristic Of Deterministic Swapping (MODS)
(Ni˜no et al., 2011) is a local serach strategy that ex-
plores the Feasible Solution Space of a Combinatorial
Problem supported in a data structure named Multiob-
jective Deterministic Finite Automata (MDFA) (Ni˜no
et al., 2010). A MDFA is a data structure that al-
lows the representation of the feasible solution space
of a Combinatorial Problem. Formally, a MDFA is
defined as follows:
M = {Q, Σ,δ,Q
0
,F(X)} (5)
Where Q represents all the set of states of the Au-
tomata (Feasible Solution Space),Σ is the input alpha-
bet that is used for δ (transition function) to explore
the feasible solution space of a Combinatorial Prob-
lem, in other words δ perturbs the solutions for find-
ing news, Q
0
contains the Initial set of States (Initial
Solutions) and F(X) are the Objectives to optimize.
The main algorithm set the MDFA to the Com-
binatorial Problem and explores it using a search di-
rection based in the elitist set of solutions (Q
). The
elitist solutions are solutions that, when were found,
dominated at least one solution from Q
φ
.
The template algorithm of MODS is defined as
follow:
Step 1. Create the initial set of solutions Q
0
using a
heuristic relative to the problem to solve.
Step 2. Set Q
φ
as Q
0
.
Step 3. Select a random state q Q
φ
or q Q
Step 4. Explore the Neighborhood of q using δ and
Σ. Add to Q
φ
the solutions found that are not domi-
nated and add to Q
those solutions that dominated at
least one element from Q
φ
.
Step 5. Check stop condition, go to 3.
3 EVOLUTIONARY
METAHEURISTIC OF
DETERMINISTIC SWAPPING
EMODS, Evolutionary Metaheuristic of Determinis-
tic Swapping, is a framework that allows the Multiob-
jective Optimization of Combinatorial Problems. Its
framework is based on MODS template therefore its
steps are the same: create Initial Solutions, Improve
ICORES 2012 - 1st International Conference on Operations Research and Enterprise Systems
400
the Solutions (Optional) and Execute the Core Algo-
rithm.
Alike MODS, Q
0
has the Initial Solutions (states)
of the Combinatorial Problem. Each state has a vector
solution. Those are created using the well known Nest
Neighbor Heuristic. Hence, a new function is created
based in the Weighted Sum Metric therefore a weight
is assigned to each Objective Function of the problem
(This is a classic manner for multiobjective optimiza-
tion(Pretorius and Helberg, 2004)) as follows:
F(X)=
n
i=1
α
i
· f
i
(X) (6)
Subject to
n
i=1
α
i
= 1 (7)
Where n is the number of objective functions. The
weights (α
i
) values are randomly assigned to each
function. Once this step has been concluded, the Nest
Neighbor Heuristic is applied to (1) for creating the
Initial Solutions. The Core Algorithm is defined as
follows:
Step 1. Set θ as the maximum number of iterations,
β as the maximum number of state selected in each
iteration, ρ as the maximum number of perturbations
by state and Q
φ
as Q
0
Step 2. Selection. Randomly select a state q Q
φ
or q Q
Step 3. Mutation. Set N as the new solutions found
as result of perturbing q. Add to Q
φ
and Q
according
to the next equations:
Q
φ
= Q
φ
{q}
6 r Q
φ
/q r
(8)
(Q
= Q
{q})
r Q
φ
/r q
(9)
Remove the states with dominated solutions for each
set.
Step 4. Crossover. Randomly select states from Q
φ
and Q
. Generate a random value k, cross the solu-
tions in a k-position as can be seen in figure 2.
Step 5. Check stop condition, go to 3.
One of the most important steps in the EMODS
algorithm is step 4. There, the algorithm applies an
Evolutionary Strategy based in the crossover step of
Genetic Algorithms for avoiding Local Optimums as
can be seen in 2. Due to the crossover is not always
made in the same point (the k value is randomly gen-
erated in each state analyzed) the variety of solutions
found are diverse avoiding local solutions.
Figure 2: Crossover step from EMODS. Cross in the k
th
-
position.
4 EXPERIMENTAL STUDIES
4.1 Experimental Settings
4.1.1 Test Instances and Parameters
EMODS was tested using the Three Objective Trav-
eling Salesman Problem (3-TSP). Formally, TSP is
defined as follows:
min
n
i=1
n
j=1
C
ij
· X
ij
(10)
Subject to:
n
j=1
X
ij
= 1,i = 1, . . . , n (11)
n
j=1
X
ij
= 1, j = 1,...,n (12)
iκ
jκ
X
ij
|κ| 1, κ {1, ... , n} (13)
X
ij
= 0,1i, j (14)
Where C
ij
is the cost of the path X
ij
and κ is any
nonempty proper subset of the cities 1,...,m. (10) is
the objective function. The goal is the optimization of
the overall cost of the tour. (11), (12) and (14) fulfills
the constrain of visiting each city only once. Lastly,
Equation (13) set the subsets of solutions, avoiding
cycles in the tour.
The test was made using well known Three - Ob-
jective Traveling Salesman Problem (3TSP) instances
from from TSP LIB(Heidelberg, ). The instances
contains problems of 100 cities. Each city is rep-
resented as a point in the space, so the distance is
computed using the euclidean distance between each
pair of points. Each algorithm was run 10 times, the
best nondominated solutions were selected for each of
one. The true solution was constructed using the best
nondominated solutions of all the sets.
4.1.2 Algorithms in Comparison
EMODS was compared against four MOSA algo-
rithm, the algorithms are: Multiobjective Simulated
EMODS: A NOVEL EVOLUTIONARY METAHEURISTIC BASED IN THE AUTOMATA THEORY FOR THE
MULTIOBJECTIVE OPTIMIZATION OF COMBINATORIALS PROBLEMS
401
Annealing (CMOSA), Ulungu Multiobjective Sim-
ulated Annealing (UMOSA), Search Multiobjective
Simulated Annealing (SMOSA) and Evolutionary
Multiobjective Simulated Annealing (EMOSA). The
most good-perfomance of them is for Evolutionary
Multiobjective Simulated Annealing as can be seen
in(Li and Landa-Silva, 2008).
4.2 Experimental Results
4.2.1 Performance Metrics
There are metrics that allow measuring the quality of
a set of optimal solutions and the performance of an
Algorithm(Jingyu et al., 2007). Most of them use two
Pareto Fronts. The first one is PF
true
and it refers to
the real optimal solutions of a combinatorial problem.
The second is PF
know
and it represents the optimal so-
lutions found by an algorithm.
Generation of Nondominated Vectors (GNDV).
It measures the number of Nondominated Solutions
generated by an algorithm.
GNDV = |PF
know
| (15)
A higher value for this metric is desired.
Generational Distance (GD). This metric mea-
sures the distance between PF
know
and PF
true
. It al-
lows to determinate the error rate of a set of solutions
relative to the real solutions.
GD =
1
|PF
know
|
·
|PF
know
|
i=1
d
i
!
(1/p)
(16)
Where d
i
is the smallest Euclidean distance be-
tween the solution i of FP
know
and the solutions of
FP
true
. p is the dimension of the combinatorial prob-
lem, it means the number of objective functions.
Inverse Generational Distance (IGD). This is an-
other distance measurement between FP
know
and
FP
true
IGD =
1
|PF
true
|
·
|PF
know
|
i=1
d
i
!
(17)
Where d
i
is the smallest Euclidean distance between
the solution i of PF
know
and the solutions of PF
true
.
Spacing (S). It measures the range variance of
neighboring solutions in PF
know
S =
1
|PF
know
| 1
2
·
|PF
know
|
i=1
d d
i
2
!
(1/p)
(18)
Where d
i
is the smallest Euclidean distance between
the solution i of PF
know
and the rest of solutions of
PF
know
. d is the mean of all d
i
. p is the dimension of
the combinatorial problem.
A value closer to 0 for this metric is desired. A
value of 0 means that all the solutions are equidistant.
4.2.2 Results Analysis
Figures 3 and 4 show a graphical comparison be-
tween EMODS Pareto Front and the rest of Com-
pared Algorithms Pareto Fronts for the instances
KROABC100 and KROBCD100 respectively. In ad-
dition, in tables 3 and 4 is measured the performance
of the algorithms for each mentioned instance respec-
tively.
Figure 3: Graphical comparison between EMODS Pareto
Front and the rest of Algorithms Pareto Fronts for the
KROABC100 instance.
Figure 4: Graphical comparison between EMODS Pareto
Front and the rest of Algorithms Pareto Fronts for the
KROBCD100 instance.
In the first instance, as can be seen in the figure 5,
most of the times, the EMODS solutions dominated
the compared algorithms solutions. It can be corrob-
orate in table 3 where EMODS generated 34077 so-
lutions in its Pareto Front having the lowest GD only
of 0,05 (5%). In the second case, EMODS had the
best performance as can be seen in figure 6 and Table
4. EMODS generated 34824 solutions in its Pareto
Front having a distance from the Real Pareto Front
ICORES 2012 - 1st International Conference on Operations Research and Enterprise Systems
402
Table 1: Measuring algorithms performance for the
KROABC100 instance with multi - objective optimization
metrics.
GVND SPACING GD IGD
SMOSA 1095 0,0599588 26,5312636 36172,0838
CMOSA 1817 0,04035959 14,3588074 29172,6591
UMOSA 2564 0,03498623 3,4396281 3333,40773
EMOSA 3194 0,03144919 2,38276567 2482,34369
EMODS 34077 0,01365555 0,05108237 129,865643
Table 2: Measuring algorithms performance for the
KROBCD100 instance with multi - objective optimization
metrics.
GVND SPACING GD IGD
SMOSA 1097 0,05749539 25,2829385 33752,4241
CMOSA 1795 0,04647346 14,1155524 28168,342
UMOSA 2472 0,03581236 3,46956584 3227,63056
EMOSA 3143 0,03150265 2,3144672 2321,81408
EMODS 34824 0,01307551 0,04979096 131,915227
Table 3: Average measuring algorithms performance for the
KROABC100 and KROBCD100 instances with multi - ob-
jective optimization metrics.
GVND SPACING GD IGD
SMOSA 1097 0,05749539 25,2829385 33752,4241
CMOSA 1795 0,04647346 14,1155524 28168,342
UMOSA 2472 0,03581236 3,46956584 3227,63056
EMOSA 3143 0,03150265 2,3144672 2321,81408
EMODS 34824 0,01307551 0,04979096 131,915227
only of 0.4 (4%).
Lastly, the metrics values in the table 5 are aver-
aged. It can be seen the superiority of EMODS solu-
tions having the lowest error distance (GD) value of
4.6%. In addition, in all the cases, the values of the
metrics applied to EMODS solutions are distant from
the others and those show the best performance.
5 CONCLUSIONS
A novel metaheuristic named EMODS was proposed.
EMODS is an Evolutionary Metaheuristic to solve
Combinatorial Problems. It is based in the Natu-
ral Selection Theory for avoiding Local Optimums.
Besides, it is based in MODS algorithm to repre-
sent the feasible solution space under the Automata
Theory. Hence, a search direction is used to ex-
plore the feasible solution space (Multiobjective De-
terministic Finite Automata). The proposed algo-
rithm was tested against metaheuristic from the spe-
cialized literature using well known instances from
the TSPLIB. EMODS showed the best performance
in all the metrics worked and in some cases the error
rate of EMODS was 4%.
REFERENCES
Fung, R., Tang, J., and Zhang, J. (2009). A multi-depot
vehicle routing problem with weight-related costs. In
Computers Industrial Engineering, 2009. CIE 2009.
International Conference on, pages 1028 –1033.
Glover, F. and Laguna, M. (1997). Tabu Search. Kluwer
Academic Publishers, Norwell, MA, USA.
Heidelberg, U. O. Tsplib - office research group dis-
crete optimization - university of heidelberg. http://
comopt.ifi.uni-heidelberg.de/software/TSPLIB95/.
Hu, B. and Raidl, G. (2008). Solving the railway traveling
salesman problem via a transformation into the classi-
cal traveling salesman problem. In Hybrid Intelligent
Systems, 2008. HIS ’08. Eighth International Confer-
ence on, pages 73 –77.
Jingyu, Y., Chongguo, L., Zhi, W., Lei, D., and Demin, S.
(2007). Diversity metrics in multi-objective optimiza-
tion: Review and perspective. In Integration Tech-
nology, 2007. ICIT 07. IEEE International Conference
on, pages 553–557.
Li, H. and Landa-Silva, D. (2008). Evolutionary multi-
objective simulated annealing with adaptive and com-
petitive search direction. In Evolutionary Com-
putation, 2008. CEC 2008. (IEEE World Congress
on Computational Intelligence). IEEE Congress on,
pages 3311 –3318.
Lim, A. and Wang, F. (2005). Multi-depot vehicle routing
problem: a one-stage approach. Automation Science
and Engineering, IEEE Transactions on, 2(4):397
402.
Ni˜no, E. D., Ardila, C., Donoso, Y., and Jabba, D. (2010). A
novel algorithm based on deterministic finite automa-
ton for solving the mono-objective symmetric travel-
ing salesman problem. International Journal of Arti-
ficial Intelligence, 5(A10):101 – 108.
Ni˜no, E. D., Ardila, C., Donoso, Y., Jabba, D., and Barrios,
A. (2011). Mods: A novel metaheuristic of determin-
istic swapping for the multi objective optimization of
combinatorials problems. Computer Technology and
Application, 2(4):280 – 292.
Oberlin, P., Rathinam, S., and Darbha, S. (2009). A trans-
formation for a heterogeneous, multiple depot, multi-
ple traveling salesman problem. In American Control
Conference, 2009. ACC ’09., pages 1292 –1297.
Pretorius, W. and Helberg, A. (2004). Application of
an adapted evaluation process using numerical amp;
qualitative weighted sum techniques. In AFRICON,
2004. 7th AFRICON Conference in Africa, volume 1,
pages 367 –372 Vol.1.
Sauer, J. and Coelho, L. (2008). Discrete differential evolu-
tion with local search to solve the traveling salesman
problem: Fundamentals and case studies. In Cyber-
netic Intelligent Systems, 2008. CIS 2008. 7th IEEE
International Conference on, pages 1 –6.
EMODS: A NOVEL EVOLUTIONARY METAHEURISTIC BASED IN THE AUTOMATA THEORY FOR THE
MULTIOBJECTIVE OPTIMIZATION OF COMBINATORIALS PROBLEMS
403
Wang, S.-Q. and Xu, Z.-Y. (2009). Ant colony algorithm
approach for solving traveling salesman with multi-
agent. In Information Engineering, 2009. ICIE ’09.
WASE International Conference on, volume 1, pages
381 –384.
Wang, Y. and Lang, M. (2008). Study on the model and
tabu search algorithm for delivery and pickup ve-
hicle routing problem with time windows. In Ser-
vice Operations and Logistics, and Informatics, 2008.
IEEE/SOLI 2008. IEEE International Conference on,
volume 1, pages 1464 –1469.
Yong-fa, Q. and Ming-yang, Z. (2004). Research on a
new multiobjective combinatorial optimization algo-
rithm. In Robotics and Biomimetics, 2004. ROBIO
2004. IEEE International Conference on, pages 187
–191.
ICORES 2012 - 1st International Conference on Operations Research and Enterprise Systems
404