ssp, mst, msf
-
Upload
kyunghyun-nam -
Category
Documents
-
view
236 -
download
0
Transcript of ssp, mst, msf
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 1/165
Graduate School of Information, Production and Systems, Waseda University
6. Network Design Problems
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 2/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 2
6. Network Design Problems Genetic Algorithms (GAs) are one of the most powerful and broadly applicable
stochastic search and optimization techniques based on principles from evolution
theory (Holland, 1976):
Michalewicz, Z. : Genetic Algorithm + Data Structure = Evolution Programs, 2nd ed.,
Springer-Verlag, New York, 1994
Gen, M. & R. Cheng: Genetic Algorithms & Engineering Design, John Wiley & Sons,
New York, 1997.
Recent advances in evolutionary computation have made it possible to solvesuch practical network optimization problems:
Ali, M. & F. Kamoun : “ Neural Networks for Shortest Path Computation and Routing
in Computer Networks”, IEEE Trans. on Neural Networks, vol.4, pp.941-954, 1993.
Perfetti, R. : “Optimization Neural Network for Solving Flow Problems”, IEEE Trans.
on Neural Network, Vol.6, No.5, pp.1287-1291, 1995. Gen, M. & K. Ida: Neural Networks and Optimization with Mathematica, Kyoritsu
Shuppan, 1998 in Japanese.
Ahn, C. W., R. Ramakrishna, C. Kang & I. Choi: “Shortest Path Routing Algorithm
using Hopfield Neural Network”, Electronic Letter, Vol.37, No.19, pp.1176-1178, 2001.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 3/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 3
6. Network Design Problems In the past few years, the genetic algorithms community has turned much of its
attention toward the optimization of network design problems:
Munakata, T. & D. J. Hashier : “A genetic algorithm applied to the maximum flow problem”, Proc.
of the 5th Inter. Conf. on Genetic Algorithms, San Francisco, pp.488-493, 1993.
Gen, M. & R. Cheng: Genetic Algorithms and Engineering Design, John Wiley & Sons, New York,
1997.
Munetomo, M., Y. Takai & Y. Sato: “A migration Scheme for the Genetic Adaptive routing
Algorithm”, Proc. of IEEE Int. Conf. Systems, Man, and Cybernetics, pp.2774-2779, 1998.
Inagaki, J., M. Haseyama & H. Kitajima: “A Genetic Algorithm for Determining Multiple Routes
and Its Applications”, Proc. of IEEE Int. Symp. Circuits and Systems, pp.137-140, 1999.
Gen, M. & R. Cheng: Genetic Algorithms and Engineering Optimization, John Wiley & Sons, New
York, 2000.
Gen, M., R. Cheng & S.S. Oren: "Network design techniques using adapted genetic algorithms",
Advances in Engineering Software, Vol.32, pp.731-744, 2001.
Ahn, C.W. & R. Ramakrishna: “A Genetic Algorithm for Shortest Path Routing Problem and the
Sizing of Populations”, IEEE Trans. on Evol. Comput., Vol.6, No.6, pp.566-579, 2002.
Zhou, G. & M. Gen: “A Genetic Algorithm Approach on Tree-like Telecommunication Network
Design Problem”, J. of Operational Research Society , Vol. 54, No. 3, pp.248-254, 2003.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 4/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 4
vBNS Backbone Network Maphttp://www.mci.com/index.jsp
vBNS: very high speed Backbone Network Services
vBNS: very high speed Backbone Network Services
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 5/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 5
vBNS Logical Network Maphttp://www.mci.com/index.jsp
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 6/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 6
6. Network Design Problems
1. Shortest Path Problem (SPP)
2. Maximum Flow (MXF) Problem
3. Minimum Cost Flow (MCF) Problem
4. Bicriteria Network Design Problem (BNP)
5. Multi-criteria Network Design Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 7/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 7
6. Network Design Problems
1. Shortest Path Problem (SPP)
1.1 Basic Concept of Shortest Path Problem1.2 Application of Shortest Path Problem
1.3 Methods for solving SPP
1.4 Genetic Approach for solving SPP
1.4.1 Reviewing Encoding Methods
1.4.2 Priority-based Genetic Algorithm1.4.3 Genetic Operators
1.5 Numerical Examples
2. Maximum Flow (MXF) Problem
3. Minimum Cost Flow (MCF) Problem
4. Bicriteria Network Design Problem (BNP)
5. Multi-criteria Network Design Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 8/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 8
1. Shortest Path Problem (SPP)
SPP is perhaps the simplest of all network designproblems.
For this problem, the object is to find a path of
minimum cost (or length) from a specified source
node s to another specified sink node t, assuming
that each arc (i, j)∈ A has an associated cost (or
length) cij.
1
2
3
5
4
6
7
8
9
10
36
27
13
18
12
16
23
11
12
38
20
32
15
13
24
s t
1 11
2
3
5
4
6
7
8
9
10
36
27
13
18
12
16
23
11
12
38
20
32
15
13
24
s t
1
2
3
5
4
6
7
8
9
10
36
27
13
18
12
16
23
11
12
38
20
32
15
13
24
s t
1 1
i j cij
1 2 36
1 3 27
2 4 18
3 2 13
3 5 123 6 23
4 7 11
4 8 32
5 4 16
6 7 12
6 9 387 8 20
8 9 15
8 10 24
9 10 13
Data table of example network
i jcij
1.1 Basic Concept of Shortest Path Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 9/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 9
1. Shortest Path Problem (SPP)
1.1 Basic Concept of Shortest Path Problem
Directed graph G =(V , A)where V is a set of nodes, A is a set of links.
cij is a cost associated with each arc(i, j)
Source node: node 1
Destination node: node n
Indicator variable:1, if link i , j is includedinthepath
0, otherwise¿
xij
=¿ { ¿ ¿ ¿
¿
1
2
3
5
4
6
7
8
9
10
36
27
13
18
12
16
23
11
12
38
20
32
15
13
24
s t
1 11
2
3
5
4
6
7
8
9
10
36
27
13
18
12
16
23
11
12
38
20
32
15
13
24
s t
1
2
3
5
4
6
7
8
9
10
36
27
13
18
12
16
23
11
12
38
20
32
15
13
24
s t
1 1
i j cij
1 2 36
1 3 27
2 4 18
3 2 13
3 5 12
3 6 23
4 7 11
4 8 32
5 4 16
6 7 12
6 9 38
7 8 20
8 9 15
8 10 24
9 10 13
Data table of example network
i jcij
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 10/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 10
1. Shortest Path Problem (SPP)
1.1 Basic Concept of Shortest Path Problem
SPP can be formulated as follows:
),,2,1,(1or 0
)(1
)1,,3,2(0
)1(1
11
1 1
n ji x
ni
ni
i
x x
xc z
ij
n
k
ki
n
j
ij
n
i
n
j
ijij
==
=−
−=
=
=
−
=
∑∑
∑∑
==
= =
s.t.
min
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 11/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 11
1. Shortest Path Problem (SPP)
1.2 Application of Shortest Path Problem
This basic model can be applied in many applications such as:Evans, J. R. and E. Minieka: Optimization Algorithms for Networks and Graphs.
New York: Marcel-Dkker, 1992.
Transportation Planning How to determine the route road that have prohibitive weight restriction so that the
driver can reach the destination within the shortest possible time.
Salesperson Routing Suppose that a sales person want to go to Los Angeles from Boston and stop over
in several city to get some commission. How can she determine the route?
Investment Planning How to determine the invest strategy to get an optimal investment plan.
Message routing in communication systems
The Routing algorithm computes the shortest (least cost) path between the router and all the networks of the internetwork.
It is one of the most important issues that has a significant impact on the network’sperformance.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 12/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 12
1. Shortest Path Problem (SPP)
1.2 Application of Shortest Path Problem
With the growth of the Internet, Internet Service Providers (ISPs) try to meetthe increasing traffic demand with new technology and improved utilizationof existing resources.
Routing of data packets can affect network utilization.
Packets are sent along network paths from source to destination followinga protocol.
Open Shortest Path First (OSPF) is the most commonly used protocol. Ericsson, M., M.G.C. Resende & P.M. Pardalos: “A Genetic Algorithm for the
Weight Setting Problem in OSPF Routing”, J. of Combinatorial Optimization,No.6, pp.299–333, 2002.
OSPF is designed for exchanging routing information within a large or verylarge internetwork.
The biggest advantage of OSPF is that it is efficient. OSPF requires very little network overhead even in very large internetworks.
The biggest disadvantage of OSPF is its complexity. OSPF requires proper planning and is more difficult to configure and
administer.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 13/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 13
1. Shortest Path Problem (SPP)
1.2 Application of Shortest Path Problem
OSPF uses a Shortest Path Routing (SPR) algorithm to compute routes inthe routing table.
The SPR algorithm computes the shortest (least cost) path between the router
and all the networks of the internetwork.
As the size of the link state database increases:
Memory requirements and route computation times increase.
Genetic Algorithm (GA) approaches to the SPR problem in OSPF.
Ahn, C.W. & R. Ramakrishna: “A Genetic Algorithm for Shortest Path Routing
Problem and the Sizing of Populations”, IEEE Trans. on Evol. Comput., Vol.6,
No.6, pp.566-579, 2002.
Lin, L., M. Gen & R. Cheng: “Priority-based Genetic Algorithm for Shortest Path
Routing Problem in OSPF”, Proc. of 3rd Inter. Conf. on Information and
Management Sciences, Dunhuang, China, June 5-10, 2004.
The objective of this research considers the quality of solution (path optimality)
within the shortest route computation times.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 14/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 14
1. Shortest Path Problem (SPP)
1.3 Methods for Solving SPP
Dijkstra Shortest Path Algorithm Dijkstra, E. W.: "A Note on Two Problems in Connection with Graphs", Numerische Math., No.1, pp.269-271, 1959.
Dijkstra's algorithm can be implemented efficiently by storing the graphin the form of adjacency lists and using a heap as priority queue toimplement the Extract-Min function.
Computes shortest paths in a graph with non-negative edge weights. Bellman-Ford Algorithm
Bellman-Ford algorithm computes single-source shortest paths in aweighted graph (where some of the edge weights may be negative).
Bellman-Ford is usually used only when there are negative edgeweights.
Floyd-Warshall Algorithm Floyd-Warshall algorithm is an algorithm to solve the all pairs shortest
path problem in a weighted, directed graph by multiplying anadjacency-matrix representation of the graph multiple times.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 15/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 15
1. Shortest Path Problem (SPP)
1.4 Genetic Approach for Solving SPP
How to encode a path in a network is critical for designing aGA.
Special difficulties:
a path contains variable number of nodes.
a random sequence of edges usually does not correspond to a
path.
Path 1 : 1→2→4→8→10
Objective function value : z =110
Path 2 : 1→2→4→7→8→10
Objective function value : z=109
Path 3 : 1→3→5→4→7→8→10
Objective function value : z=110
1
2
3
5
4
6
7
8
9
10
36
27
13
18
12
16
23
11
12
38
20
32
15
13
24
s t
1 11
2
3
5
4
6
7
8
9
10
36
27
13
18
12
16
23
11
12
38
20
32
15
13
24
s t
1
2
3
5
4
6
7
8
9
10
36
27
13
18
12
16
23
11
12
38
20
32
15
13
24
s t
1 1
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 16/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 16
1. Shortest Path Problem (SPP)
1.4.1 Reviewing Encoding Methods
How to encode a solution of the problem into a chromosome is a key issuefor GAs.
For the nonstring coding approach, three critical issues emergedconcerning with the encoding and decoding between chromosomes andsolutions: The feasibility of a chromosome
Feasibility refers to the phenomenon of whether a solution decoded from a
chromosome lies in the feasible region of a given problem. The legality of a chromosome
Legality refers to the phenomenon of whether a chromosome represents asolution to a given problem.
The illegality of chromosomes originates from the nature of encodingtechniques.
Repairing techniques are usually adopted to convert an illegal chromosome to
a legal one. The uniqueness of mapping
The mapping from chromosomes to solutions (decoding) may belong to one of the following three cases: (a) 1-to-1 mapping; (b) n-to-1 mapping; (c) 1-to-n mapping.
The 1-to-1 mapping is the best one among three cases And 1-to-n mapping is the most undesired one.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 17/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 17
1.4.1 Reviewing Encoding Methodsa. Priority-based Chromosome (Cheng & Gen, 1997)
Cheng & Gen proposed a priority-based encoding method for solving resource-
constrained project scheduling problem (rcPSP) first. And also adopted this method for solving SPP in 1997.
Cheng, R. & M. Gen: “Resource Constrained Project Scheduling Problem using Genetic Algorithm”, Inter. J. of
Intelligent Auto. and Soft Comput., Vol.3, pp.273-286, 1997.
Gen, M., R. Cheng & D. Wang: “Genetic Algorithms for Solving Shortest Path Problems”, Proc. of IEEE Int. Conf.
on Evol. Comput., Indianapolis, Indiana, pp.401-406, 1997.
They adopted an indirect approach: The path is generated by sequential node appending procedure with beginning from
the specified node 1 and terminating at the specified node n.
At each step, there are usually several nodes available for consideration.
They gave each node a priority with a random mechanism and add the one with the
highest priority into path.
As we know, a gene in a chromosome is characterized by two factors:
locus, i.e., the position of gene located within the structure of chromosome,
allele, i.e., the value which the gene takes.
In the priority-based encoding method, the position of a gene is used to represent node ID and its value is used to
represent the priority of the node for constructing a path among candidates. A path can be uniquely determined from
this encoding.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 18/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 18
1.4.1 Reviewing Encoding Methods
a. Priority-based Chromosome (Cheng & Gen, 1997)
Example: An example of generated chromosome and its decoded path
as follows:
Advantage:
Any permutation of the encoding corresponds to a path (legality).
Most existing genetic operators can be easily applied to the encoding.
Any path has a corresponding encoding (completeness); any point in
solution space is accessible for genetic search.
Disadvantage:
At some case, n-to-1 mapping may occur for the encoding.
1 4
3 6
7
s t
2 5
1 1
node ID :1 2 3 4 5 6 7
priority :2 1 6 4 5 3 7
1 43 7 path :
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 19/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 19
1.4.1 Reviewing Encoding Methodsb. Variable-length Chromosome (Munemoto et al., 1998)
Munemoto et. al . (1998) proposed a variable-length encoding method for network routing problems in a wired or wireless environment. Ahn et. al .
(2002) also used the encoding method for solving the shortest path routing
(SPR) problem.
Munetomo, M., Y. Takai & Y. Sato: “A migration Scheme for the Genetic Adaptive routing
Algorithm”, Proc. of IEEE Int. Conf. Systems, Man, and Cybernetics, pp.2774-2779, 1998.
Ahn, C.W. & R. Ramakrishna: “A Genetic Algorithm for Shortest Path Routing Problem and theSizing of Populations”, IEEE Trans. on Evol. Comput., Vol.6, No.6, pp.566-579, 2002.
The proposed encoding method consists of sequences of positive integers
that represent the IDs of nodes through which a path passes.
Each locus of the chromosome represents an order of a node (indicated by the gene of the
locus) in a path.
The length of the chromosome is variable, but is should not exceed the maximum length n,
where n is the total number of nodes in the network, since it never needs more than n number of
nodes to form a path.
The gene of the first locus encodes the source node, and the gene of second locus is randomly
or heuristically selected from the nodes connected with the source node.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 20/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 20
1.4.1 Reviewing Encoding Methodsb. Variable-length Chromosome (Munemoto et al., 1998)
Example: An example of generated chromosome and its decoded path as follows:
Advantage:
The mapping from any chromosome to solution (decoding) belongs to 1-
to-1 mapping (uniqueness).
Theoretically, convergence performance is better than the priority-based
encoding method.
Disadvantage:
In general, the genetic operators may generate infeasible chromosomes
(illegality) that violate the constraints, generating loops in the paths.
Repairing techniques are usually adopted to convert an illegal
chromosome to a legal one.
1 4
3 6
7
s t
2 5
1 1
1 43 7 path :
locus :1 2 3 4
node ID :1 3 4 7
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 21/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 21
1.4.1 Reviewing Encoding Methods
c. Fixed-length Chromosome (Inagaki et al., 1999)
Inagaki et al. (1999) proposed a fixed-length encoding method
determining multiple routes in routing applications.
Inagaki, J., M. Haseyama & H. Kitajima: “A Genetic Algorithm for Determining
Multiple Routes and Its Applications”, Proc. of IEEE Int. Symp. Circuits and
Systems, pp.137-140, 1999.
The proposed method are sequences of integers and each gene
represents the node ID through which it passes.
To encode a route from node 1 to node n, put i in the jth locus of the
chromosome.
This process is reiterated from the specified node 1 and terminating at
the specified node n.
If the route does not pass through a node x, select one node randomly
from the set of nodes which are connected with node x, and put it in
the xth locus.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 22/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 22
1.4.1 Reviewing Encoding Methods
c. Fixed-length Chromosome (Inagaki et al., 1999)
Example: An example of generated chromosome and its decoded path as follows:
Advantage:
Any path has a corresponding encoding (completeness).
Any point in solution space is accessible for genetic search.
Any permutation of the encoding corresponds to a path (legality) using thespecial genetic operators.
Disadvantage: At some case, n-to-1 mapping may occur for the encoding.
Furthermore the probability of occurrence of n-to-1 mapping is higher than thepriority-based encoding method.
In the special genetic operator phase, some offspring may generate newchromosomes that resemble the initial chromosomes in fitness, therebyretarding the process of evolution.
1 4
3 6
7
s t
2 5
1 1
1 43 7 path :
locus : 1 2 3 4 5 6 7
node ID :3 1 4 7 2 4 6
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 23/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 23
1.4.1 Reviewing Encoding Methods
Compared with the Performance of Different Encoding Methods:
Variable-length encoding method Convergence performance is best than others.
However, the genetic operators may generate infeasible chromosomes
(illegality).
Repairing techniques have to be adopted to convert an illegal
chromosome to a legal one. For the computation times, variable-length
encoding method may be slow in several large network design problems.
Fixed-length encoding method
n-to-1 mapping may occur for the encoding.
The special genetic operators have to been adopted; thereby some
offspring may generate new chromosomes that resemble the initial
chromosomes in fitness.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 24/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 24
1.4.2 Priority-based Genetic Algorithm
procedure 1: Priority-based Encoding
input: number of nodes n
output: chromosome v k
begin
for j=1 to n // step 0 v
k ( j )← j ;
for i=1 to // step 1
repeat
j ←random[1, n];
l ←random[1, n];
until l ≠ j
swap (v k ( j ), v
k (l ));
output the chromosome v k ; // step 2
end
procedure 1: Priority-based Encoding
input: number of nodes n
output: chromosome v k
begin
for j=1 to n // step 0 v
k ( j )← j ;
for i=1 to // step 1
repeat
j ←random[1, n];
l ←random[1, n]; until l ≠ j
swap (v k ( j ), v
k (l ));
output the chromosome v k ; // step 2
end
Priority-based Encoding Method
2/n
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 25/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 25
1.4.2 Priority-based Genetic Algorithm
procedure 2: One Path Growthinput: number of nodes n, chromosome v
k ,
the set of nodes S i with all nodes adjacent to node i.
output: path P k
begin
initial source node i ←1, P k
←φ ; // step 0
while S i ≠φ do // step 1
select l from S i with the highest priority;
if v k (l )≠0 then
v k (l )=0;
P k
←P k
∪{ x il
};
i ←l ;
else S i ←S
i \{l }
end
output the complete path P k ; // step 2
end
Decoding Method
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 26/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 26
1.4.2 Priority-based Genetic Algorithm
i j cij
1 2 36
1 3 27
2 4 18
3 2 13
3 5 123 6 23
4 7 11
4 8 32
5 4 16
6 7 12
6 9 387 8 20
8 9 15
8 10 24
9 10 13
Data table of example network
Path: 1→3→6→7→8→10
Objective function value : z =106
priority: v ( j )
node ID: j
91108526437
10987654321
Chromosome:
Illustration of Priority-based GA
18
1
2
3
5
4
6
7
8
9
10
36
27
13
12
16
23
11
12
38
20
32
15
13
24
s t1 1
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 27/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 27
It operates on two parents (chromosomes) at a time and generatesoffspring by combining both chromosomes’ features. In network design problem, crossover plays the role of exchanging each partial route of two
chosen parents in such a manner that the offspring produced by the crossover represents. In this study, the nature of the priority-based encoding is a kind of permutation
representation. Generally, this representation will yield illegal offspring by one-point crossover or other simple
crossover operators.
During the past decade, several crossover operators have been proposed for permutation representation, such as: Partial-mapped crossover (PMX)
Goldberg, D. & R. Lingle, Alleles: “loci and the traveling salesman problem”, Proc. of the 1st Inter.Conf. on GA, pp.154-159, 1985.
Order crossover (OX):
Davis, L. : “Applying adaptive algorithms to domains”, Proc. of the Inter. Joint Conf. on
Artificial Intelligence, pp.1162-164, 1985. Position-based crossover (PX)
Davis, L. : “Applying adaptive algorithms to domains”, Proc. of the Inter. Joint Conf. on Artificial Intelligence, pp.1162-164, 1985.
Cycle crossover (CX)
Oliver, I. & J. Holland: “A study of permutation crossover operators on the traveling salesmanproblem, Euro. J. of OR , vol.26, pp.187-210, 1986.
Heuristic crossover , and so on.
1.4.3 Genetic Operators --- Crossover
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 28/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 28
1.4.3 Genetic Operators --- Crossover Partial-Mapped Crossover (PMX) PMX was proposed by Goldberg and Lingle.
Goldberg, D. & R. Lingle, Alleles: “loci and the traveling salesman problem”, Proc.
of the 1st Inter. Conf. on GA, pp.154-159, 1985.
PMX can be viewed as an extension of two-point crossover for binary
string to permutation representation.
It uses a special repairing procedure to resolve the illegitimacy caused
by the simple two-point crossover.step 1 : select the substring at random
step 2 : exchange substrings between
step 3 : determine mapping relationship
step 4 : legalize offspring with mapping
relationship
parent 1: 1 7 2 3 4 6 5 8
parent 2: 4 6 3 5 7 1 8 2
substring selected
parent 1: 1 7 3 5 7 6 5 8
parent 2: 4 6 2 3 4 1 8 2
2 3 4
3 5 7 74
532
↔
↔↔
offspring 1: 1 4 3 5 7 6 2 8
offspring 2: 7 6 2 3 4 1 8 5
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 29/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 29
1.4.3 Genetic Operators --- Crossover
Order Crossover (OX)
OX was proposed by Davis.Davis, L. : “Applying adaptive algorithms to domains”, Proc. of the Inter.
Joint Conf. on Artificial Intelligence, pp.1162-164, 1985.
It can be viewed as a kind of variation of PMX with a different
repairing procedure.
parent 1: 1 7 2 3 4 6 5 8
parent 2: 4 6 3 5 7 1 8 2
offspring : 6 5 2 3 4 7 1 8
substring selected
Fig. 6.1 Illustration of the OX operator.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 30/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 30
1.4.3 Genetic Operators --- Crossover
Position-based Crossover (PX)
PX was proposed by Syswerda.Davis, L. : “Applying adaptive algorithms to domains”, Proc. of the Inter.
Joint Conf. on Artificial Intelligence, pp.1162-164, 1985.
It is essentially a kind of uniform crossover for permutation
representation together with a repairing procedure.
It also can be viewed as a kind of variation of OX in which thenodes are selected inconsecutively.
parent 1: 1 7 2 3 4 6 5 8
parent 2: 4 6 3 5 7 1 8 2
offspring : 3 7 5 1 4 6 2 8
Fig. 6.2 Illustration of the PX operator.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 31/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 31
1.4.3 Genetic Operators --- Crossover
However, in all of above approaches: the mechanism of the crossover is not the same as that of the
conventional one-point crossover.
Some offspring may generate new chromosomes that are notpossible to succeed the character of the parents.
thereby retarding the process of evolution.
We proposed a new crossover operator, Weight MappingCrossover (WMX). WMX can be viewed as an extension of one-point crossover for
permutation representation.
As one-point crossover:
Two chromosomes (parents) would be to choose a random cut-point. Generate the offspring by using segment of own parent to the left of
the one-cut point
Then remapping the right segment that base on the weight of other parent of right segment .
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 32/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 32
1.4.3 Genetic Operators --- Crossover
Weight Mapping Crossover (WMX)
];:1[//]:1[
];:1[//]:1[
]);[(sorting][]);[(sorting][
];[][
];[][
1
;
];,1[random
,
122
211
22
11
22
11
21
21
n pv pv' v
n pv pv' v
s s s s
i pvi s
i pvi s
l i
pnl
n p
' v',v
nv ,v
+←
+←
⋅←⋅⋅←⋅
+←
+←
=
−←
←
tofor
begin
:output
:input
CrossoverMappingWeight:procedure
end
output
thenif
tofor
thenif
tofor
tofor
;,
];[][][][
1
];[][
][][
1
1
21
22
12
11
21
' v' v
j si p' v j si p' v
l j
j si p' v
j si p' v
l j
l i
←+
=+
=
←+
=+
=
=
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 33/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 33
1.4.3 Genetic Operators --- Crossover
Weight Mapping Crossover (WMX) As shown Fig., first we choose a random cut-point p.
calculate l that is the length of right segments of chromosomes, where n isnumber of nodes in the network.
Then get mapping relationship by sorting the weight of the right segments s1[∙] and
s2[∙].
As one-point crossover, generate the offspring v1
’, v2
’ by exchange substrings
between parents v1, v
2; legalize offspring with mapping relationship.
6354712 parent 1 :
4156273 parent 2 :
cut-point
415
635653
541
5146273offspring 2 :
5364712offspring 1 :
step 1: select a cut-point
step 2: mapping the weight of the right segment
step 3: generate offspring with mapping relationship
1 43 7 parent 1 :
1 42 7 parent 2 :
offspring 1 :
1 42 7offspring 2 :
5
1 43 75
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 34/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 34
1.4.3 Genetic Operators --- Mutation It is relatively easy to produce some mutation operators for
permutation representation.
During the past decade, several mutation operators havebeen proposed for permutation representation, such as: Inversion
Insertion
Displacement
Swap mutation.
Insertion Mutation Selects a gene at random and inserts it in a random position as
follows:
6354712 parent :
select a gene at random
6347152offspring :
insert it in a random position
1 43 7 parent :
offspring : 1 4 7
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 35/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 35
1.4.3 Genetic Operators --- Immigration
The trade-off between exploration and exploitation in
serial GAs for function optimization is a fundamentalissue.
If a GA is biased towards exploitation:
highly fit members are repeatedly selected for recombination.
Although this quickly promotes better members, the population
can prematurely converge to a local optimum of the function.
If a GA is biased towards exploration:
Large numbers of schemata are sampled which tends to
inhibit premature convergence.
Unfortunately, excessive exploration results in a large number of function evaluations, and defaults to random search in the
worst case.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 36/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 36
1.4.3 Genetic Operators --- Immigration
To search effectively and efficiently, a GA must maintain a balance
between these two opposing forces.
Michael, C.M., C.V. Stewart & R.B. Kelly: “Reducing the Search Time of A Steady
State Genetic Algorithm using the Immigration Operator”, Proc. of IEEE Int. Conf.
on Tools for AI San Jose, CA, pp.500-501, 1991.
Michael et. al. (1991) proposed an immigration operator which, for certain types
of functions, allows increased exploration while maintaining nearly the same level
of exploitation for the given population size.
Immigration operator
step 1: The algorithm is modified to include immigration, with each
generation generated.
step 2: Evaluate μ random members (μ, called the immigration rate).
step 3: Replace the μ worst members of the population with the μ random
members. This study experimentally examines the immigration operator, and
present the effectiveness of this approach for solving network
design problems in next section.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 37/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 37
1.4.3 Genetic Operators --- Selection Selection operators: two basic types of selection scheme
used commonly in current practice.
Proportionate selection: picks out chromosomes based on their fitness values relative to the fitness of the other chromosomes inthe population. Roulette wheel selection
Stochastic remainder selection
Stochastic universal selection
Ordinal-based selection: upon their rank within the population.The chromosomes are ranked according to their fitness values. Tournament selection
selection
Truncation selection
Linear ranking selection
In this study, the roulette wheel selection, a type of Proportionate selection, is adopted.
),( λ µ
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 38/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 38
1.4.4 Overall Procedure
procedure: Priority-based GA for Shortest Path Probleminput: network data (V, A, C ), GA parameters
output: best shortest path
begin
t 0;
initialize P (t ) by priority-based encoding;fitness eval (P );
while (not termination condition) do
crossover P (t ) to yield C (t ) by weight mapping crossover ;
mutation P (t ) to yield C (t ) by insertion mutation;
immigration operation to yield C (t )
fitness eval (C );
select P (t+1) from P (t ) and C (t ) by roulette wheel selection;
t t + 1;
end
output best shortest path;
end
GA Procedure for Shortest Path Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 39/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 39
1.5 Numerical Examples
Test Problems: For examining the effect of different encoding methods, we applied Ahn
et al’s method and priority-based encoding method to the 6 testproblems: Ahn, C.W. & R. Ramakrishna: “A Genetic Algorithm for Shortest Path Routing Problem
and the Sizing of Populations.” IEEE Trans. Evol. Comput., Vol.6, No.6, pp.566-579,2002.
OR-Notes. [Online]. Available: http://mscmga.ms.ic.ac.uk/jeb/or/orweb.html
Using the following parameter specifications. Population size: popSize =20
Crossover probability: pC =0.70
Mutation probability: pM =0.50
Immigration rate: μ=3
Maximum generation: maxGen =1000
Terminating condition: 100 generations with same fitness.
Each solution is compared with Dijkstra’s algorithm that provides areference point (optimal solution).
Each algorithm was applied to each test problem 20 times (i.e., 20 runs)using different initial populations.
All the simulations were performed with Java on Pentium 4 processor (1.5-GHz clock).
1
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 40/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 40
1
1
1.5 Numerical Examples
The first numerical example, presented by Ahn et al’s was adopted.
The problem comprises 20 nodes and 49 arcs. It is given as follows:
(Ahn, C.W. & R. Ramakrishna: “A Genetic Algorithm for Shortest Path Routing Problemand the Sizing of Populations”, IEEE Trans. on Evol. Comput., Vol.6, No.6, pp.566-
579, 2002.)
Fig.6.3 Example of the first numerical example
1 5 N i l E l
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 41/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 41
Convergence property of each algorithm for a Fixed Network
With 20 Nodes(Ahn, C.W. & R. Ramakrishna: “A Genetic Algorithm for Shortest Path Routing Problem and
the Sizing of Populations”, IEEE Trans. on Evol. Comput., Vol.6, No.6, pp.566-579, 2002.)
Dijkstra’s AlgorithmMunemoto’s AlgorithmInagaki’s AlgorithmAhn’s Algorithm
2.5
2.0
1.5
1.0
0.5
0 2 4 6 8 10
Generations
Objective
Function
Values
1.5 Numerical Examples
1 5 N i l E l
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 42/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 42
240
260
240
260
1.5 Numerical Examples
Convergence property of Ahn et al.’s algorithm and proposed
algorithm for a Fixed Network With 20 Nodes
Fig. 6.4 Convergence property of Ahn et al.’s algorithm and proposed algorithm.
et al.
1 5 N i l E l
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 43/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 43
Comparison with results(Ahn, C.W. & R. Ramakrishna: “A Genetic Algorithm for Shortest Path Routing Problem and
the Sizing of Populations”, IEEE Trans. on Evol. Comput ., Vol.6, No.6, pp.566-579, 2002.)
Inagaki’s Algorithm
Munemoto’s Algorithm
Ahn’s Algorithm
15 20 25 30 35 40 45 50
1200
1000
800
600
400
200
0
The Number of Nodes
Objective
FunctionValue
1.5 Numerical Examples
1 5 N i l E l
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 44/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 44
1.5 Numerical Examples
Discussion of the Results:
The quality of solution with different genetic operators is investigated in Table 1.
The path optimality is defined in all test problems, by Alg.6 (WMX+Insertion+ Immigration)that the GA finds the global optimum (i.e., the shortest path).
The path optimality is defined in #1, #2 test problems, by Alg.5 (WMX+Swap+ Immigration),The near optimal result is defined in other test problems.
By Alg.1 ~ Alg.4, the path optimality is not defined. Since the number of possible alternativesbecome to very large in test problems, the population be prematurely converged to a localoptimum of the function.
Table 6.1 Performance comparisons with different genetic operatorsTest Problems
(# of nodes/ # of arcs)
OptimalSolutions
Best Solutions
Alg. 1 Alg. 2 Alg. 3 Alg. 4 Alg. 5 Alg. 6
20/49 142.00 148.35 148.53 147.70 143.93 142.00 142.00
80/120 389.00 423.53 425.33 418.82 396.52 389.00 389.00
80/632 291.00 320.06 311.04 320.15 297.21 291.62 291.00
160/2544 284.00 429.55 454.98 480.19 382.48 284.69 284.00320/1845 394.00 754.94 786.08 906.18 629.81 395.01 394.00
320/10208 288.00 794.26 732.72 819.85 552.71 331.09 288.00
Alg. 1: FMX+Swap; Alg. 2: OX+Swap; Alg. 3: PX+Swap; Alg. 4: WMX+Swap;
Alg. 5: WMX+Swap+Immigration(3); Alg. 6: WMX+Insertion+Immigration(3).
1 5 N i l E l
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 45/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 45
Comparison results of Ahn’s algorithm and Proposed algorithm
Test Problems
(# of nodes/ # of arcs)
OptimalSolutions
Best Solutions CPU Times (ms) Generation Num. of Obtained best result
Ahn’s Alg. Prop. Alg. Ahn’s Alg. Prop. Alg. Ahn’s Alg. Prop. Alg.
20/49 142.00 142.00 142.00 40.60 23.37 2 9
80/120 389.00 389.00 389.00 118.50 96.80 4 4
80/632 291.00 291.00 291.00 109.50 118.50 19 10
160/2544 284.00 286.20 284.00 336.20 490.50 31 26320/1845 394.00 403.40 394.00 779.80 1062.50 44 11
320/10208 288.00 288.90 288.00 1028.30 1498.50 38 26
20/49 80/120 80/630 160/2544 320/1845 320/10208
Bestsolutions
Problem size
Table 6.2 Performance comparisons with Ahn’s algorithm and Proposed algorithm.
1.5 Numerical Examples
1 5 N i l E l
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 46/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 46
1.5 Numerical Examples
Different Parameter Settings:
Parameter Settings
( popSize / pC : pM )
Test Problems
(# of nodes/ # of arcs)
OptimalSolutions
Best Solutions CPU Times ( ms ) Generation Num. of Obtained best result
Ahn’s Alg. Prop. Alg. Ahn’s Alg. Prop. Alg. Ahn’s Alg. Prop. Alg.
10 / 0.3 : 0.1 20/49 142.00 156.20 142.00 10.42 8.37 38 27
80/120 389.00 389.00 389.00 32.80 31.10 5 1
80/632 291.00 313.20 291.00 29.40 34.40 43 16
160/2544 284.00 320.90 284.20 67.10 106.30 48 37
320/1845 394.00 478.70 394.00 120.30 250.20 68 18
320/10208 288.00 444.00 288.30 126.40 400.20 25 59
20 / 0.3 : 0.1 20/49 142.00 145.23 142.00 22.36 13.34 27 24
80/120 389.00 389.00 389.00 56.30 51.50 4 1
80/632 291.00 303.10 291.00 50.10 56.30 18 10
160/2544 284.00 298.70 284.20 122.10 181.20 44 35
320/1845 394.00 465.70 394.00 213.90 496.70 32 17
320/10208 288.00 373.10 288.60 311.00 631.10 61 35
20 / 0.7 : 0.5 20/49 142.00 142.00 142.00 40.60 23.37 6 9
80/120 389.00 389.00 389.00 118.50 96.80 1 1
80/632 291.00 291.00 291.00 109.50 118.50 19 10
160/2544 284.00 286.20 284.00 336.20 490.50 31 26
320/1845 394.00 403.40 394.00 779.80 1062.50 44 11
320/10208 288.00 288.90 288.00 1028.30 1498.50 38 26
Table 6.3 Performance comparisons with different parameter settings
1 5 N i l E l
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 47/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 47
Different Parameter Settings with Ahn’s algorithm and Proposed algorithm
Parameter Settings
( popSize / pC : pM )
Probability of obtaining the optimal solutions
Ahn’s Alg. Proposed Alg.
10 / 0.3 : 0.1 16.67% 66.67%
20 / 0.3 : 0.1 16.67% 66.67%
30 / 0.3 : 0.1 33.33% 83.33%
50 / 0.3 : 0.1 50.00% 100.00%
100 / 0.3 : 0.1 33.33% 100.00%
10 / 0.7 : 0.5 33.33% 83.33%20 / 0.7 : 0.5 50.00% 100.00%
30 / 0.7 : 0.5 50.00% 100.00%
50 / 0.7 : 0.5 83.33% 100.00%
100 / 0.7 : 0.5 83.33% 100.00%
1.5 Numerical Examples
1 5 N i l E l
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 48/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 48
1.5 Numerical Examples Simulation (# of nodes: 100, # of arcs: 859)
6 Network Design Problems
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 49/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 49
6. Network Design Problems
1. Shortest Path Problem (SPP)
2. Maximum Flow (MXF) Problem2.1 Basic Concept of Maximum Flow Problem
2.2 Application of Maximum Flow Problem
2.3 Methods for solving MXF Problem
2.4 Genetic Approach for solving MXF Problem
2.4.1 Genetic Representation2.4.2 Genetic Operators
2.5 Numerical Examples
1. Minimum Cost Flow (MCF) Problem
2. Bicriteria Network Design Problem (BNP)
3. Multi-criteria Network Design Problem
2 Maximum Flow (MXF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 50/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 50
[Online]. Available: http://www-b2.is.tokushima-u.ac.jp/
~ikeda/suuri/maxflow/Maxflow.shtml.en
MXF is in a sense a complementary model to SPP.
MXF seeks a feasible solution that sends the maximum
amount of flow from a specified source node s to another
specified sink node t.
If we interpret uij as the maximum flow rate of arc (i, j),
MXF identifies the maximum steady-state flow that the
network can send from node s to node t per unit time.
i j uij
1 2 60
1 3 601 4 60
2 3 30
2 5 40
2 6 30
3 4 30
3 6 50
3 7 30
4 7 40
5 8 60
6 5 20
6 8 30
6 9 40
6 10 30
7 6 20
7 10 40
8 9 30
8 11 60
9 10 30
9 11 50
10 11 50
Data table of example network
i juij
2.1 Basic Concept of Maximum Flow Problem
60
1
2
3
4
5
6
7
8
9
10
11
s t60
60
30
30
20
20
30
30
60
50
50
40
30
50
30
40
60
30
40
30
40
f f
2. Maximum Flow (MXF) Problem
2 Maximum Flow (MXF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 51/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 51
2. Maximum Flow (MXF) Problem
2.1 Basic Concept of Maximum Flow Problem
Directed graph G =(V , A)
where V is a set of nodes, A is a set of links.
uij is a capacity associated with each link(i, j)
Source node: node 1
Destination node: node n
i juij
i j uij
1 2 60
1 3 601 4 60
2 3 30
2 5 40
2 6 30
3 4 30
3 6 50
3 7 30
4 7 40
5 8 60
6 5 20
6 8 30
6 9 40
6 10 30
7 6 20
7 10 40
8 9 30
8 11 60
9 10 30
9 11 50
10 11 50
Data table of example network
60
1
2
3
4
5
6
7
8
9
10
11
s t60
60
30
30
20
20
30
30
60
50
50
40
30
50
30
40
60
30
40
30
40
f f
2 Maximum Flow (MXF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 52/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 52
2. Maximum Flow (MXF) Problem
2.1 Basic Concept of Maximum Flow Problem
MXF problem can be formulated as follows:
f z =max
−
=−∑ ∑= =
f
f
x xn
j
n
k
kiij 0 t.s.1 1
)(
)1,,3,2(
)1(
ni
ni
i
=−=
=
A jiu x ijij ∈≤≤ ),(,0
0≥ f
2 Maximum Flow (MXF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 53/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 53
2. Maximum Flow (MXF) Problem
2.2 Application of Maximum Flow Problem
This basic MXF model can be applied in many applicationssuch as:
Ahuj, R. K., T. L. Magnanti & J. B. Orlin: Network Flows. Prentice-Hall, Upper Saddle River, NJ, 1993.
Scheduling on Uniform Parallel Machines The feasible scheduling problem, described in the preceding paragraph, is a
fundamental problem in this situation and can be used as a subroutine for more general scheduling problems, such as the maximum lateness problem,the (weighted) minimum completion time problem, and the (weighted)maximum utilization problem.
Distributed Computing on a Two-Processor Computer Distributed computing on a two-processor computer concerns assigning
different modules (subroutines) of a program to two processors in a way thatminimizes the collective costs of interprocessor communication and
computation..
Tanker Scheduling Problem A steamship company has contracted to deliver perishable goods between
several different origin-destination pairs. Since the cargo is perishable, thecustomers have specified precise dates (i.e., delivery dates) when theshipments must reach their destinations..
2 Maximum Flow (MXF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 54/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 54
2. Maximum Flow (MXF) Problem
2.3 Methods for solving MXF Problem
Ford-Fulkerson Algorithm It works by finding a flow augmenting path in the graph. By adding the
flow augmenting path to the flow already established in the graph, the
maximum flow will be reached when no more flow augmenting paths
can be found in the graph.
Maximum Flow Algorithm An incremental algorithm for max-flow problem that tries to find the
max-flow in the network as an edge is deleted or inserted in the
network, is presented.
It has also been shown that other cases of a unit change can be
considered as a special case of insertion and deletion of an edge inthe network.
2 Maximum Flow (MXF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 55/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 55
2.4 Genetic Approach for solving MXF Problem
Munakata, T. and D. J. Hashier : “A genetic algorithm applied to the maximum flowproblem,” Proc. of 5th Int. Conf. on Genetic Algorithms, pp. 488-493, 1993.
The maximum flow problem appears to be more challenging in applying GAs
than many other common graph problems (e.g., shortest path, minimum
spanning tree)
Its unique characteristic:
A flow at each edge can be anywhere between zero and its flow capacity, i.e.,
it has more "freedom" to choose.
In many other problems, selecting an edge may mean to simply add a fixed
distance.
In the maximum flow problem, two conditions must be satisfied:
The flow at each edge must be between zero and its flow capacity.
At each vertex, the incoming flow and outgoing flow must balance.
2. Maximum Flow (MXF) Problem
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 56/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 56
2.4 Genetic Approach for solving MXF Problem
2.4.1 Genetic Representation
procedure 1: Priority-based Encodinginput: number of nodes n
output: chromosome v k
begin
for j=1 to n // step 0
v k ( j )← j ;
for i=1 to // step 1
repeat
j ←random[1, n];
l ←random[1, n];
until l ≠ j
swap (v k ( j ), v
k (l ));
output the chromosome v k ; // step 2
end
procedure 1: Priority-based Encodinginput: number of nodes n
output: chromosome v k
begin
for j=1 to n // step 0
v k ( j )← j ;
for i=1 to // step 1
repeat
j ←random[1, n];
l ←random[1, n];
until l ≠ j
swap (v k ( j ), v
k (l ));
output the chromosome v k ; // step 2
end
2/n
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 57/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 57
2.4 Genetic Approach for solving MXF Problem
The decoding procedure is a two-stage process.
First stage: the path is generated by one-path growth procedure
It is given in procedure 2
With beginning from the specified node 1 and terminating at the specified
node n. At each step, add the one with the highest priority into path.
procedure 2: One-path Growthinput: number of nodes n, chromosome v
k ,
the set of nodes S i with all nodes adjacent to node i.
output: path P k
step 0: the source node i ←1, P k ←φ
step 1: if S i =φ , goto step 3; otherwise, continue.
step 2: select l from S i with the highest priority, and go back to step 1.
if v k (l )≠0 then
v k (l )=0;
P k ←P
k ∪{ x
il };
i ←l ;
else v k (l )=0
step 3: output the complete path P k ;
}...,,,,{ ,,,1 132211 mm l l l l l l l k x x x x P
−
=
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 58/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 58
2.4 Genetic Approach for solving MXF Problem
The decoding procedure is a two-stage process.
Second stage: overall paths are generated by overall pathsgrowth procedure
For a given path, we can calculate its flow f k
By removing the used capacity from uij of each arc, we have a
new network with the new flow capacity ũij. With the one-path growth procedure (procedure 2), we can obtain
the second path.
By repeating this procedure we can obtain the maximum flow for
the given chromosome till no new network can be defined in this
way.
It is given in procedure 3.
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 59/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 59
2.4 Genetic Approach for solving MXF Problem
procedure 3: Overall-path Growth
input: network data (V, A, U ), chromosome v k , the set of nodes S
i with all nodes adjacent to node i
output: number of paths Lk , the flow f i k of each path, i ∈Lk
step 0: number of paths l ←0
step 1: if S 1=φ , go to step 7; otherwise, l ← l +1, continue.
step 2: the implementation of path P l k growth is based on procedure 2. Select the sink node a of path p
l k .
step 3: if the sink node a=n, continue; otherwise, perform the set of nodes S i update as follows, return to
step 1.
step 4: calculate the flow f l k of the path P
l k .
step 5: perform the flow capacity u ij of each arc update. Make a new flow capacity ũ ij as follows:
step 6: if the flow capacity ũ ij =0, perform the set of nodes S
i update which the node j adjacent to node i .
step 7: output number of paths Lk ← l -1, the flow f
i k of each path, i ∈L
k .
iaS S ii ∀−← },{
}),(|min{1 lk ij
k
l
k
l P jiu f f ∈+← −
}),(min{~ k l ijijij P jiuuu ∈−←
0~&),( ,}{ =∈−← ij
k
l ii u P ji j s s
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 60/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 60
2.4 Genetic Approach for solving MXF Problem
Illustration of Priority-based GA
node ID :1 2 3 4 5 6 7 8 9 10 11
priority :2 1 6 4 11 9 8 10 5 3 7
i j uij
1 2 60
1 3 601 4 60
2 3 30
2 5 40
2 6 30
3 4 30
3 6 50
3 7 30
4 7 40
5 8 60
6 5 20
6 8 30
6 9 40
6 10 30
7 6 20
7 10 40
8 9 30
8 11 60
9 10 30
9 11 50
10 11 50
Data table of example network
60
1
2
3
4
5
6
7
8
9
10
11
s t60
60
30
30
20
20
30
30
60
50
50
40
30
50
30
40
60
30
40
30
40
Chromosome:
f f
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 61/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 61
Illustration of Priority-based GA
node ID :1 2 3 4 5 6 7 8 9 10 11
priority :2 1 6 4 11 9 8 10 5 3 7
Chromosome:
60
s
360
s
3
f
k i S i l P k S 1 f k
1 0 1
1 2, 3, 4 3 1, 3
2 4, 6, 7 6 1, 3, 6
3 5, 8, 9, 10 5 1, 3, 6, 5
4 8 8 1, 3, 6, 5, 8
5 9, 11 11 1, 3, 6, 5, 8, 11 2, 3, 4 20
2 0 1
1 2, 3, 4 3 1, 3
2 4, 6, 7 6 1, 3, 6
3 8, 9, 10 8 1, 3, 6, 8
4 9, 11 11 1, 3, 6, 8, 11 2, 3, 4 50
k : number of paths
i : start node
S i : the set of nodes
l : sink node
P k : the k th pathS 1 : the set of nodes with all nodes
adjacent to node 1
f k : maximum possible flow
2.4 Genetic Approach for solving MXF Problem
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 62/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 62
Illustration of Priority-based GA
Chromosome:
60
s
360
s
3
f
k i S i l P k S 1 f k
3 0 1
1 2, 3, 4 3 1, 3
2 4, 7 7 1, 3, 7
3 6, 10 6 1, 3, 7, 6
4 9, 10 9 1, 3, 7, 6, 9
5 10, 11 11 1, 3, 7, 6, 9, 11 2, 4 60
4 0 11 2, 4 4 1, 4
2 7 7 1, 4, 7
3 6, 10 6 1, 4, 7, 6
4 9, 10 9 1, 4, 7, 6, 9
5 10, 11 11 1, 4, 7, 6, 9, 11 2,4 70
k : number of paths
i : start node
S i : the set of nodes
l : sink node
P k : the k th pathS 1 : the set of nodes with all nodes
adjacent to node 1
f k : maximum possible flow
node ID :1 2 3 4 5 6 7 8 9 10 11
priority :2 1 6 4 11 9 8 10 5 3 7
2.4 Genetic Approach for solving MXF Problem
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 63/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 63
Illustration of Priority-based GA
Chromosome:
60
s
360
s
3
f
k i S i l P k S 1 f k
5 0 1
1 2, 4 4 1, 4
2 7 7 1, 4, 7
3 10 10 1, 4, 7, 10
4 11 11 1, 4, 7, 10, 11 2 100
6 0 1
1 2 2 1, 2
2 3, 5, 6 5 1, 2, 5
3 8 8 1, 2, 5, 8
4 9, 11 11 1, 2, 5, 8, 11 2 110
k : number of paths
i : start node
S i : the set of nodes
l : sink node
P k : the k th pathS 1 : the set of nodes with all nodes
adjacent to node 1
f k : maximum possible flow
node ID :1 2 3 4 5 6 7 8 9 10 11
priority :2 1 6 4 11 9 8 10 5 3 7
2.4 Genetic Approach for solving MXF Problem
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 64/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 64
Illustration of Priority-based GA
Chromosome:
60
s
360
s
3
f
k i S i l P k S 1 f k
7 0 1
1 2 2 1, 2
2 3, 5, 6 5 1, 2, 5
3 8 8 1, 2, 5, 8
4 9 9 1, 2, 5, 8, 9
5 10, 11 11 1, 2, 5, 8, 9, 11 2 140
8 0 11 2 2 1, 2
2 3, 6 6 1, 2, 6
3 9, 10 9 1, 2, 6, 9
4 10 10 1, 2, 6, 9, 10
5 11 11 1, 2, 6, 9, 10, 11 160
k : number of paths
i : start node
S i : the set of nodes
l : sink node
P k : the k th pathS 1 : the set of nodes with all nodes
adjacent to node 1
f k : maximum possible flow
node ID :1 2 3 4 5 6 7 8 9 10 11
priority :2 1 6 4 11 9 8 10 5 3 7
2.4 Genetic Approach for solving MXF Problem
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 65/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 65
Illustration of Priority-based GA
Objective function value : z =160
Chromosome:
i j uij
1 2 60
1 3 60
1 4 60
2 3 30
2 5 40
2 6 30
3 4 30
3 6 50
3 7 30
4 7 40
5 8 60
6 5 20
6 8 30
6 9 40
6 10 307 6 20
7 10 40
8 9 30
8 11 60
9 10 30
9 11 50
10 11 50
Data table of example network
60/60
1
2
3
4
5
6
7
8
9
10
11
s t60/60
40/60
20/20
20/20
30/30
20/30
60/60
50/50
50/50
40/40
20/30
50/50
10/3040/40
60/60
30/30
40/40
30/40
160 160
node ID :1 2 3 4 5 6 7 8 9 10 11
priority :2 1 6 4 11 9 8 10 5 3 7
i j xij / uij
2.4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 66/165
2 4 Genetic Approach for solving MXF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 67/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 67
2.4 Genetic Approach for solving MXF Problem
Mutation: The swap mutation operator was used here, in which twopositions are selected at random and their contents are swapped
as follows:
Selection: The roulette wheel approach, a type of fitness-
proportional selection, was adopted.
8 parent : 5643271 8 parent : 5643271
8offspring : 5243671 8offspring : 5243671
exchanging points
2.4.2 Genetic Operators
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 68/165
2.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 69/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 69
2.5 Numerical Examples
Test Problems: The numerical examples, presented by T. Munakata &
D.J. Hashier, was adopted.
Munakata, T. and D. J. Hashier : “A genetic algorithm applied to themaximum flow problem,” Proc. of 5th Int. Conf. on Genetic Algorithms,pp. 488-493, 1993.
Using the following parameter specifications. Population size: popSize =10
Crossover probability: pC =0.50
Mutation probability: pM =0.50
Maximum generation: maxGen =1000
Terminating condition: 100 generations with same fitness.
All the simulations were performed with Java onPentium 4 processor (1.5-GHz clock).
2.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 70/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 70
2.5 Numerical ExamplesTest Problem 1:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25 f
20
20
20
20
20
20
20
20
20
20
20
20
20
20
10
10
10
10
10
10
10
10
8
8
8
8
8
25
25
25
25
25
20
15
15
15
15
15
15
15
15
6
5
4
5
4
30
30
30
f
The first numerical example, presented by Munakata & Hashier , was
adopted. The problem comprises 25 nodes and 49 arcs. It is given asfollows:
i juij
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 71/165
2.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 72/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 72
2.5 Numerical Examples Process of Genetic Computing
f l o w
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 73/165
2.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 74/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 74
i juij
Objective function value: z=91 (optimal solution)
Generation Num. of Obtained best result: 67
Best Chromosome:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
2591 91
18/20
20/20
20/20
20/20
13/20
10/10
8/8
10/10
6/6
4/8
5/5
7/9
8/8
7/7
10/10
3/7
7/12
6/15
18/18
9/12
15/20
7/22
3/6
8/11
2/2
4/5
6/10
4/8
6/10
1/10
6/6
4/5
9/10
7/8
9/9
5/7
8/15
8/8
8/8
6/6
5/9
15/15
20/20
22/30
20/20
14/15
node ID : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
priority: 20 10 5 22 25 23 11 6 18 1 16 12 3 2 7 14 15 19 4 13 17 8 21 24 9
2.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 75/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 76/165
6. Network Design Problems
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 77/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 77
g
1. Shortest Path Problem (SPP)
2. Maximum Flow (MXF) Problem
3. Minimum Cost Flow (MCF) Problem
3.1 Basic Concept of MCF Problem
3.2 Application of MCF Problem
3.3 Methods for solving MCF Problem3.4 Genetic Approach for solving MCF Problem
3.4.1 Genetic Representation
3.4.2 Genetic Operators
3.5 Numerical Examples
4. Bicriteria Network Design Problem (BNP)
5. Multicriteria Network Design Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 78/165
3. Minimum Cost Flow (MCF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 79/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 79
3.1 Basic Concept of MCF Problem Directed graph G =(V , A)
where V is a set of nodes, A is a set of links. uij is a capacity associated with each link(i, j)
cij is unit cost associated with each link(i, j)
Source node: node 1
Destination node: node n
i j cij
uij
1 2 18 60
1 3 19 601 4 17 60
2 3 13 30
2 5 16 40
2 6 14 30
3 4 15 30
3 6 16 50
3 7 17 304 7 19 40
5 8 19 60
6 5 15 20
6 8 16 30
6 9 15 40
6 10 18 30
7 6 15 20
7 10 13 40
8 9 17 30
8 11 18 60
9 10 14 30
9 11 19 50
10 11 17 50
Data table of example network
i jcij , uij
18, 60
1
2
3
4
5
6
7
8
9
10
11
s t19, 60
17, 60
13, 30
15, 30
15, 20
15, 20
17, 30
14, 30
18, 60
17, 50
19, 50
16, 40
14, 30
16, 50
17, 30
19, 40
19, 60
16, 30
15, 40
18, 30
13, 40
q q
( )
3. Minimum Cost Flow (MCF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 80/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 80
( )
3.1 Basic Concept of MCF Problem
MCF problem can be formulated as follows:
∑∑= =
=n
i
n
j
ijij xc z 1 1
min
−
=−∑ ∑= =
q
q
x xn
i
n
k
kiij 0 t.s.1 1
)(
)1,,3,2(
)1(
ni
ni
i
=
−=
=
A jiu x ijij ∈≤≤ ),(,0
q: total flow valueq: total flow value
3. Minimum Cost Flow (MCF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 81/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 81
( )
3.2 Application of Minimum Cost Flow (MCF) Problem
This basic MCF model can be applied in many applicationssuch as:
Ahuj, R. K., T. L. Magnanti & J. B. Orlin: Network Flows, Prentice-Hall, Upper Saddle River, NJ, 1993.
Transportation Problem
There are a set of nodes called sources, and a set of nodes called
destinations. All arcs go from a source to a destination. There is a per-unitcost on each arc. Each source has a supply of material, and eachdestination has a demand.
It can be solved by applying Min-cost Flow Algorithm
Distribution Problem
The distribution of a product from manufacturing plants to warehouses, or
from warehouses to retailers The flow of raw material and intermediate goods through the various
machining stations in a production line
The routing of automobiles through an urban street network
The routing of calls through the telephone system.
3. Minimum Cost Flow (MCF) Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 82/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 82
( )
3.3 Methods for solving MCF ProblemAhuj, R. K., T. L. Magnanti & J. B. Orlin: Network Flows, Prentice-Hall, Upper
Saddle River, NJ, 1993.
Successive Shortest Path Algorithm
The successive shortest path algorithm maintains optimality of the solution at
every step and strives to attain feasibility.
Primal-dual Algorithm
The primal-dual algorithm for the minimum cost flow problem is similar to the
successive shortest path algorithm in the sense that it also maintains a
pseudoflow that satisfies the reduced cost optimality conditions and gradually
converts it into a flow by augmenting flows along shortest paths.
Out-of-Kilter Algorithm
The out-of-kilter algorithm, which satisfies only the mass balance constraints,
so intermediate solutions might violate both the optimality conditions and the
flow bound restrictions.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 83/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 84/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 85/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 86/165
3.4 Genetic Approach for solving MCF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 87/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 87
step 4: calculate the flow f l k and the cost c
l k of the path P
l k .
step 5: perform the flow capacity u ij
of each arc update. Make a new
flow capacity ũ ij as follows:
step 6: if the flow capacity ũ ij =0, perform the set of nodes S
i update
which the node j adjacent to node i .
step 7: output number of paths Lk ← l -1, the flow f
i k and the cost c
i k of
each path, i ∈Lk
.
k
l
n
i
n
j
k
l
k
l ij
k
l
k
l
lk ijk l k l
P ji f f ccc
P jiu f f
∈∀−+←
∈+←
∑∑= =
−−
−
),(),(
}),(|min{
1 1
11
1
}),(min{~ k
l ijijij P jiuuu ∈−←
0~&),( ,}{ =∈−← ijk
l ii u P ji j s s
3.4 Genetic Approach for solving MCF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 88/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 88
Illustration of Priority-based GA i j cij
uij
1 2 18 60
1 3 19 601 4 17 60
2 3 13 30
2 5 16 40
2 6 14 30
3 4 15 30
3 6 16 50
3 7 17 304 7 19 40
5 8 19 60
6 5 15 20
6 8 16 30
6 9 15 40
6 10 18 30
7 6 15 20
7 10 13 40
8 9 17 30
8 11 18 60
9 10 14 30
9 11 19 50
10 11 17 50
Data table of example network
i jcij , uij
18, 60
1
2
3
4
5
6
7
8
9
10
11
s t19, 60
17, 60
13, 30
15, 30
15, 20
15, 20
17, 30
14, 30
18, 60
17, 50
19, 50
16, 40
14, 30
16, 50
17, 30
19, 40
19, 60
16, 30
15, 40
18, 30
13, 40
q q
node ID :1 2 3 4 5 6 7 8 9 10 11
priority :2 1 6 4 11 9 8 10 5 3 7
Chromosome:
3.4 Genetic Approach for solving MCF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 89/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 89
Illustration of Priority-based GA
k i S i l P k S 1 f l k cl
k
1 0 1
1 2, 3, 4 3 1, 3
2 4, 6, 7 6 1, 3, 6
3 5, 8, 9, 10 5 1, 3, 6, 5
4 8 8 1, 3, 6, 5, 8
5 9, 11 11 1, 3, 6, 5, 8, 11 2, 3, 4 20 87
2 0 1
1 2, 3, 4 3 1, 3
2 4, 6, 7 6 1, 3, 6
3 8, 9, 10 8 1, 3, 6, 8
4 9, 11 11 1, 3, 6, 8, 11 2, 3, 4 30 69
k : number of paths
i : start node
S i : the set of nodes
l : sink node
P k : the k th path
S 1 : the set of nodes with all nodes
adjacent to node 1
f l k : the total flow
cl k : minimum possible cost
60
node ID :1 2 3 4 5 6 7 8 9 10 11
priority :2 1 6 4 11 9 8 10 5 3 7
Chromosome:
3.4 Genetic Approach for solving MCF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 90/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 90
Illustration of Priority-based GA
k i S i l P k S 1 f l k cl
k
3 0 1
1 2, 3, 4 3 1, 3
2 4, 7 7 1, 3, 7
3 6, 10 6 1, 3, 7, 6
4 9, 10 9 1, 3, 7, 6, 9
5 10, 11 11 1, 3, 7, 6, 9, 11 2, 4 10 85
k : number of paths
i : start node
S i : the set of nodes
l : sink node
P k : the k th path
S 1 : the set of nodes with all nodes
adjacent to node 1
f l k : the total flow
cl k : minimum possible cost
60
node ID :1 2 3 4 5 6 7 8 9 10 11
priority :2 1 6 4 11 9 8 10 5 3 7
Chromosome:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 91/165
3.4 Genetic Approach for solving MCF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 92/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 92
3.4.2 Genetic Operators
step 1 : select the substring at random
step 2 : exchange substrings between
step 3 : determine mapping relationship
parent 1: 1 7 2 3 4 6 5 8
parent 2: 4 6 3 5 7 1 8 2
substring selected
parent 1: 1 7 3 5 7 6 5 8
parent 2: 4 6 2 3 4 1 8 2
2 3 4
3 5 7 74
532
↔
↔↔
offspring 1: 1 4 3 5 7 6 2 8
offspring 2: 7 6 2 3 4 1 8 5
• Here the position-based crossover operator proposed by PMX (Partial
Mapped Crossover) (Gen-Cheng97, pp.119-125) was adopted.
• It uses a special repairing procedure to resolve the illegitimacy caused by
the simple two-point crossover as follows:
step 4 : legalize offspring with mapping
relationship
3.4 Genetic Approach for solving MCF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 93/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 93
Mutation: The swap mutation operator was used here, in which twopositions are selected at random and their contents are swapped
as follows:
Selection: The roulette wheel approach, a type of fitness-
proportional selection, was adopted.
8 parent : 5643271 8 parent : 5643271
8offspring : 5243671 8offspring : 5243671
exchanging points
3.4.2 Genetic Operators
3.4 Genetic Approach for solving MCF Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 94/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 94
procedure: Priority-based GA for Solving MCF Problem
input: network data (V, A, C, U ), GA parameters
output: best minimum cost
begin
t 0;
initialize P (t ) by priority-based encoding;
fitness eval (P );while (not termination condition) do
crossover P (t ) to yield C (t ) by partial mapped crossover ;
mutation P (t ) to yield C (t ) by swap mutation;
fitness eval (C );
select P (t+1) from P (t ) and C (t ) by roulette wheel selection;t t + 1;
end
output best minimum cost;
end
GA Procedure for solving MCF Problem
3.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 95/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 95
Test Problems: The numerical examples, presented by Munakata &
Hashier , was adopted.
Munakata, T. and D. J. Hashier : “A genetic algorithm applied to themaximum flow problem,” Proc. of 5th Int. Conf. on Genetic Algorithms,pp. 488-493, 1993.
Using the following parameter specifications. Population size: popSize =10
Crossover probability: pC =0.50
Mutation probability: pM =0.50
Maximum generation: maxGen =1000
Terminating condition: 100 generations with same fitness.
All the simulations were performed with Java onPentium 4 processor (1.5-GHz clock).
3.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 96/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 96
Test Problem 1:
The first numerical example, presented by Munakata & Hashier , was
adopted. The problem comprises 25 nodes and 49 arcs. It is given asfollows:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25q
10, 20
13, 20
32, 20
135, 20
631, 20
8, 20
6, 20
7, 20
7, 20
7, 20
5, 20
14, 20
4, 20
14, 20
10, 10
35, 10
3, 10
33, 10
7, 10
7, 10
3, 10
10, 10
13, 8
15, 8
11, 8
5, 8
3, 8
11, 25
8, 25
35, 25
14, 25
12, 25
34, 20
10, 15
4, 15
9, 15
11, 15
12, 15
9, 15
14, 15
5, 15
10, 6
15, 533, 4
4, 5
13, 4
10, 30
2, 30
3, 30
q
i jcij , uij
3.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 97/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 97
Total flow value q = 70
Objective function value: z= 6969
Generation Num. of Obtained best result:863
Best Chromosome:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25q=70
10, 18/20
13, 14/20
32, 20/20
135, 14/20
631, 4/20
6, 10/20
7, 20/20
7, 18/20
14, 7/20
4, 15/20
10, 10/10
35, 10/10
3, 10/10
7, 4/10
3, 10/10
10, 10/10
13, 8/8
15, 8/8
3, 8/8
11, 10/25
8, 20/25
14, 20/25
10, 15/15
9, 15/15
11, 15/15
12, 7/15
9, 3/15
14, 15/15
10, 5/6
15, 5/5
33, 4/4
4, 2/5
13, 4/4
10, 30/30
2, 20/30
3, 20/30
q=70
node ID : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
priority: 1 16 11 9 6 5 7 8 15 10 3 12 13 21 4 22 14 18 20 24 17 25 23 2 19
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 98/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 99/165
3 10/10 29 10/11
3.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 100/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 100
Total flow value q = 72
Objective function value: z=5986
Generation Num. of Obtained best result:132
Best Chromosome:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
2, 18/20
10, 14/20
33, 20/20
126, 20/20
3, 10/10
14, 8/8
13, 6/6
32, 8/8
7, 3/5
12, 9/9
33, 8/8
4, 7/7
12, 10/10
30, 3/7
7, 8/18
6, 6/12
8, 15/20
3, 5/5
11,2 /22
6, 6/6
29, 10/11
30, 6/10
34,9/10
2, 3/9
35, 6/6
14, 2/5
8, 6/10
26, 8/9
12, 6/7
31, 3/15
8,4/ 4
30, 8/8
11, 4/6
2,10/15
10, 20/20
6, 16/30
7, 20/20
9, 6/15
node ID : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
priority: 10 25 15 24 11 4 7 8 12 6 5 9 13 14 3 16 17 18 1 20 22 19 23 21 2
q=72q=72
3.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 101/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 101
Process of Genetic Computing
5900
6000
6100
6200
6300
6400
6500
6600
6700
6800
6900
7000
0 50 100 150 200 250
generation
c o s t
3.5 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 102/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 102
Simulation (# of nodes: 80, # of arcs: 857)
6. Network Design Problems
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 103/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 103
1. Shortest Path Problem (SPP)
2. Maximum Flow (MXF) Problem3. Minimum Cost Flow (MCF) Problem
4. Bicriteria Network Design Problem (BNP)
4.1 Introduction of BNP
4.2 BNP Formulation4.3 Genetic Approach for solving BNP
4.3.1 Genetic Representation
4.3.2 Decoding Method
4.3.3 Fitness Assignment
4.3.4 Genetic Operators
4.4 Numerical Examples
5. Multi-criteria Network Design Problem
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 104/165
4.1 Introduction of BNP
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 105/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 105
The problems may arise when designing:
In a communication network, find a set of links which consider the low cost
(or delay) and the high throughput (or reliability) for increasing the networkperformance. Marathe, M. V., R. Ravi, R. Sundaram, S. S. Ravi, D. J. Rosenkrantz, and H. B. Hunt:
“Bicriteria network design problems,” J. Algorithms, vol. 28, no. 1, pp. 142-171, Jul. 1998.
Yuan, D.: “A bicriteria optimization approach for robust OSPF routing,” Proc. IPOM , 2003,pp. 91-98.
Yang, H., M. Maier, M. Reisslein, and W. M. Carlyle: “A genetic algorithm-based
methodology for optimizing multiservice convergence in a metro WDM network,” J.Lightwave Technol., vol. 21, no. 5, pp. 1114-1133, May. 2003.
In a manufacturing system, the two criteria under consideration areminimizing cost and maximizing manufacturing. Raghavan, S., M. O. Ball, and V. S. Trichur: “Bicriteria product design optimization,”
Institute for Systems Research, Tech. Rep. TR 2001-8, 2001.
[Online]. Available: http://techreports.isr.umd.edu/ARCHIVE/
In a logistic system, the main drive to improve logistics productivity is theenhancement of customer services and asset utilization through a significantreduction in order cycle time (lead time) and logistics costs. Zhou, G. , H. Min, and M. Gen: “A genetic algorithm approach to the bi-criteria allocation
of customers to warehouses,” Int. J. Production Economics, vol. 86, pp. 35-45, Oct. 2003.
4.1 Introduction of BNP
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 106/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 106
The Bicriteria Network Design Problem (BNP) is known as NP-hard (Gareyand Johnson, 1979), it is not simply an extension from single objective totwo objectives.
Garey, M. and D. Johnson: Computers and Intractability: A Guide to the Theory of NP-Completeness, W.H. Freeman, New York, 1979.
In generally, we can not get the optimal solution of the problem because theseobjectives usually conflict with each other in practice. The real solutions to theproblem are a set of Pareto optimal solutions (Chankong and Haimes, 1983). Chankong, V. and Y.Y. Haimes: Multiobjective Decision Making Theory and
Methodology. North-Holland, New York, 1983. For solving the BNP, the set of efficient paths may be very large and possibly
exponential in size. Thus the computational effort required to solve it can increase exponentially
with the problem size in the worst case.
While the tractability of the problem is of importance when solving large scale
problems, the issue concerning with the size of the efficient set is important toa decision maker.
Having to evaluate a large efficient set in order to select the best one poses aconsiderable cognitive burden on decision makers. Therefore, in such cases,obtaining the entire Pareto optimal set is of little interest to decision makers.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 107/165
4.1 Introduction of BNP
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 108/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 108
The bicriteria shortest path problem is one of BNPs, which of finding a
diameter-constrained shortest path from a specified source node s to
another specified sink node t. This problem, termed the multi-objective shortest path problem
(MOSP) in the literature is NP-complete.
Warburton (1987) presented the first fully polynomial approximation
scheme (FPAS) for it.
Warburto, A.: “Approximation of Pareto optima in multiple-objective, shortest
path problems,” Operations Research, vol. 35, no. 1, pp. 70-79, 1987.
Hassin (1992) provided a strongly polynomial FPAS for the problem which
improved the running time of Warburton.
Hassin, R.: “Approximation schemes for the restricted shortest path problem,”
Math. Of Operations Research, vol. 17, no. 1, pp. 36-42, Feb. 1992.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 109/165
4.1 Introduction of BNP
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 110/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 110
In this study, we dominated BNP with more complexity cases as two criteriaproblem that maximum flow and minimum cost considered. Priority-based encoding method (Cheng and Gen, 1994) has been improved.
Cheng, R. and M. Gen: “Evolution program for resource constrained projectscheduling problem,” Proc. of Int. Conf. Evol. Comput., pp.736-741, 1994. For maximizing flow, different form other genetic representation methods, such
as path oriented encoding method, priority-based encoding method canrepresent various efficient paths by each chromosome.
Considering the characteristic of priority-based encoding method, we proposeda new crossover operator called as Weight Mapping Crossover (WMX)
Insertion mutation operator and Immigration operator (Michael et al., 1991)was adopted. Michael, C.M., C.V. Stewart and R. B. Kelly: “Reducing the search time of a
steady state genetic algorithm using the immigration operator”, Proc. IEEE Int.Conf. Tools for AI, San Jose, CA, pp.500-501, 1991.
These methods provide a search capability that results in improved quality of solution and enhanced rate of convergence.
For ensure the population diversity in MOGA, Adaptive Weight Approach (AWA) which is one of weighted-sum approach, was adopted. Gen, M. and R. Cheng: Genetic Algorithms and Engineering Optimization, John
Wiley & Sons, New York, 2000. Their elements represent that weights are adjusted adaptively based on the
current generation to obtain search pressure toward the positive ideal point.
4.2 BNP Formulation
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 111/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 111
4.2 BNP Formulation In this study, we present a mathematical programming formulation of the bicriteria
network design model including MXF model and MCF model. Different from the generic BNP, the problem’s efficient set of paths may be very
large, possibly exponential in size. Thus the computational effort required to solve it
can increase exponentially with the problem size in the worst case.
In a network with flow capacities and costs on the arcs, BNP is to determine
both the maximum possible flow z 1 and minimum cost z
2 in the same time, from
a source to a sink.
0
),(, 0
)(
)1,,3,2(0
)1(
t.s.
min
max
1 1
1 1
2
1
≥
∈∀≤≤
=−
−=
=
=−
=
=
∑ ∑
∑∑
= =
= =
f
A jiu x
ni f
ni
i f
x x
xc z
f z
ijij
n
j
n
k
kiij
n
i
n
j
ijij
4.3.1 Genetic Representation
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 112/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 112
Priority-based encoding method
procedure 1: Priority-based Encoding
input: number of nodes n
output: chromosome v k
begin
for j=1 to n // step 0
v k ( j )← j ;for i=1 to // step 1
repeat
j ←random[1, n];
l ←random[1, n];
until l ≠ j
swap (v k ( j ), v
k (l ));
output the chromosome v k ; // step 2
end
procedure 1: Priority-based Encoding
input: number of nodes n
output: chromosome v k
begin
for j=1 to n // step 0
v k ( j )← j ;for i=1 to // step 1
repeat
j ←random[1, n];
l ←random[1, n];
until l ≠ j
swap (v k ( j ), v
k (l ));
output the chromosome v k ; // step 2
end
2/n
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 113/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 114/165
4.3.2 Decoding Method
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 115/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 115
procedure 3: Overall-path Growth
input: network data (V, A, C, U ), chromosome v k
, the set of nodes S i
with all nodes adjacent to node i
output: number of paths Lk
, the flow f i k and the cost c
i k of each path,
i ∈Lk
step 0: number of paths l ←0step 1: if S
1=φ , go to step 7; otherwise, l ← l +1, continue.
step 2: the implementation of path P l k growth is based on procedure 2.
Select the sink node a of path pl k .
step 3: if the sink node a=n, continue; otherwise, perform the set of
nodes S i update as follows, return to step 1.iaS S ii ∀−← },{
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 116/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 117/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 118/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 119/165
Illustration of Decoding Method
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 120/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 120
k i S i l P k S 1 z 1k z 2
k
7 0 11 2 2 1, 2
2 3, 5, 6 5 1, 2, 5
3 8 8 1, 2, 5, 8
4 9 9 1, 2, 5, 8, 9
5 10, 11 11 1, 2, 5, 8, 9, 11 2 140 12790
8 0 1
1 2 2 1, 2
2 3, 6 6 1, 2, 6
3 9, 10 9 1, 2, 6, 9
4 10 10 1, 2, 6, 9, 10
5 11 11 1, 2, 6, 9, 10, 11 160 14350
k : number of paths
i : start node
S i : the set of nodes
l : sink node
P k : the k th path
S 1 : the set of nodes with all nodes
adjacent to node 1
z 1k : maximum possible flow
z 2k : minimum possible cost
18 6018 60
Chromosome:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 121/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 122/165
4.3.3 Fitness Assignment
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 123/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 123
step 2: The adaptive weight for objective 1 and objective 2 are calculated by
the following equation:
step 3: Calculate the fitness value for each individual.
minmax
minmax
22
2
11
1
1
1
z z w
z z w
−=
−=
( ) popSizek
L
z cw z f wveval
k
L
i
k
i
k
i
k
k
∈∀+−−
=∑
= ,)()(
)( 1
min
22
min
11
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 124/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 125/165
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 126/165
4.3.5 GA Procedure for BNP
GA Procedure for BNP
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 127/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 127
procedure: Priority-based GA for BNP
input: network data (V, A, C, U ), GA parametersoutput: Pareto optimal solution E (t )
begin
t 0;
initialize P (t ) by priority-based encoding;
objectives z 1(P ), z 2(P );
create Pareto E (P );
fitness eval (P ) by adaptive weight approach;
while (not termination condition) do
crossover P (t ) to yield C (t ) by weight mapping crossover ;
mutation P (t ) to yield C (t ) by insertion mutation;
immigration operation to yield C (t ) ;
objectives z 1(C ), z 2(C );
update Pareto E (P, C );fitness eval (P, C ) by adaptive weight approach;
select P (t+1) from P (t ) and C (t ) by roulette wheel selection;
t t + 1;
end
output Pareto optimal solution E (t );
end
GA Procedure for BNP
4.4 Numerical Examples Test Problems:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 128/165
Soft Computing Lab.WASEDA UNIVERSITY , IPS
128
Test Problems: The numerical examples, presented by Munakata & Hashier , was
adopted.
Munakata, T. and D. J. Hashier: “A genetic algorithm applied to the maximum flowproblem,” Proc. of 5th Int. Conf. on Genetic Algorithms, pp. 488-493 , 1993.
Using the following parameter specifications.
Population size: popSize =20
Crossover probability: pC =0.40 Mutation probability: pM =0.60
Maximum generation: maxGen =1000
Terminating condition: 100 generations with same fitness.
All the simulations were performed with Java on Pentium 4
processor (1.5-GHz clock).
4.4 Numerical ExamplesTest Problem 1:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 129/165
Soft Computing Lab.WASEDA UNIVERSITY , IPS
129
Test Problem 1:
The first numerical example, presented by Munakata & Hashier , was
adopted. The problem comprises 25 nodes and 49 arcs. It is given asfollows:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25 f
10, 20
13, 20
32, 20
135, 20
631, 20
8, 20
6, 20
7, 20
7, 20
7, 20
5, 20
14, 20
4, 20
14, 20
10, 10
35, 10
3, 10
33, 10
7, 10
7, 10
3, 10
10, 10
13, 8
15, 8
11, 8
5, 8
3, 8
11, 25
8, 25
35, 25
14, 25
12, 25
34, 20
10, 15
4, 15
9, 15
11, 15
12, 15
9, 15
14, 15
5, 15
10, 6
15, 533, 4
4, 5
13, 4
10, 30
2, 30
3, 30
f
i jcij , uij
4.4 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 130/165
Soft Computing Lab.WASEDA UNIVERSITY , IPS
130
z 1 z 2 z 1 z 2 z 1 z 2
4 300 30 2470 72 7703
5 345 33 2786 73 8382
8 600 38 2926 75 9762
10 696 40 3046 78 11799
12 993 43 3274 80 13147
15 1001 47 3674 82 14531
18 1226 52 4074 85 17115
20 1568 56 4830 87 17941
21 1629 59 5406 88 19254
23 1833 66 6575 89 19333
28 2178 69 7145 90 20007
Table 6.4 The Pareto optimal solutions of test problem 1
Gen, M., L. Lin & R. Cheng: “Bicriteria Network Optimization Problem using Priority-
based Genetic Algorithm,” IEEJ Trans. on Elect., Info. & Sys., Oct. 2004.
4.4 Numerical Examplesideal point: z 1=90, z 2=300
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 131/165
Soft Computing Lab.WASEDA UNIVERSITY , IPS
131Fig. 6.5 The Pareto optimal solutions of test problem 1
p 1 , 2
z 1=66, z 2=6575
0 20 40 60 80 100
flow
cost
0 -
-5000 -
-10000 -
-15000 -
-20000 -
- 25000 -
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 132/165
4.4 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 133/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 133
z 1 z 2 z 1 z 2 z 1 z 2 z 1 z 22 52 32 1633 54 3872 73 6944
8 248 34 1909 55 3990 74 7192
10 340 36 1937 58 4146 75 7402
15 495 38 2077 61 4671 76 7532
18 692 40 2485 63 5153 78 7847
19 1012 41 2581 65 5463 80 9228
20 1111 43 2731 66 5704 82 10395
21 1220 47 3080 67 6323 83 12508
25 1292 49 3302 68 6422 85 12610
26 1406 51 3551 71 6537 86 13151
27 1457 52 3739 72 6748 91 16752
28 1475
Gen, M., L. Lin & R. Cheng: “Bicriteria Network Optimization Problem using Priority-
based Genetic Algorithm,” IEEJ Trans. on Elect., Info. & Sys., Oct. 2004.
Table 6.5 The Pareto optimal solutions of test problem 2
4.4 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 134/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 134
Pareto optimal solutionideal oint
0 20 40 60 80 100
flow
cost
ideal point: z 1=91, z 2=52
z 1=61, z 2=4671
0 -
-2000 -
-4000 -
-6000 -
-8000 -
-10000 -
-12000 -
-14000 -
-16000 -
-18000 -
Fig. 6.6 The Pareto optimal solutions of test problem 2
4.4 Numerical Examples Simulation (# of nodes: 25, # of arcs: 56)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 135/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 135
6. Network Design Problems
1 Shortest Path Problem (SPP)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 136/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 136
1. Shortest Path Problem (SPP)
2. Maximum Flow (MXF) Problem
3. Minimum Cost Flow (MCF) Problem
4. Bicriteria Network Design Problem (BNP)
5. Multi-criteria Network Design Problem
5.1 Introduction of Multi-criteria Network Design Problem
5.2 Reviewing Solution Approaches for MNP
5.3 Numerical Examples
5. Multi-criteria Network Design Problem (MNP)
With the information superhighway fast becoming a reality
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 137/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 137
With the information superhighway fast becoming a reality,the problem of designing networks capable of
accommodating multimedia (both audio and video) traffic ina multicast (simultaneous transmission of data to multipledestinations) environment has come to assume paramountimportance Chow, C.-H.: “On multicast path finding algorithms,” Proceedings of IEEE
INFOCOM , pp.1274-1283, 1991. Frank, A., L. Wittie, and A. Bernstein: “Multicast communication in networkcomputers,” IEEE Software, Vol. 2, No. 3, pp. 49-61,1985.
Kadaba, B. and J. Jaffe: “Routing to multiple destinations in computer networks,” IEEE Transactions on Communications, Vol. COM-31, pp. 343-351,1983.
Kompella, V.P., J.C. Pasquale and G.C. Polyzos: “Multicasting for multimedia applications,” Proceedings of IEEE INFOCOM, 1992.
Kompella, V.P., J.C. Pasquale and G.C. Polyzos: “Multicast routing for multimedia communication,” IEEE/ACM Transactions on Networking , pp.286-292, 1993.
5.1 Introduction of MNP
Network design problems where even one cost measure must
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 138/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 138
Network design problems where even one cost measure must
be minimized, are often NP-hard. But, in real-life applications, it
is often the case that the network to be built is required tominimize multiple cost measures simultaneously, with different
cost functions for each measure.
For example, in the problem of finding good multicast trees,
each edge has associated with it two edge costs: The construction cost: It is typically a measure of the amount of
buffer space or channel bandwidth used
The delay cost: It is a combination of the propagation, transmission
and queuing delays.
5.1 Introduction of MNP
Multi-criteria network design problems, with separate cost functions for
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 139/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 139
g p , peach optimization criterion, also occur naturally in Information Retrievaland VLSI designs.
Bookstein, A. & S.T. Klein: “Construction of Optimal Graphs for Bit-Vector Compression,” Proc. 13th ACM-SIGIR , vol. 16, pp. 387-400, 1990.
Zhu, Q., M. Parsa & W.W.M. Dai: “An iterative approach for delay-boundedminimum Steiner tree construction,” Technical Report UCSC-CRL-94-39, UCSanta Cruz, 1994.
With the advent of deep micron VLSI designs, the feature size has shrunk
to sizes of 0.5 microns and less. As a result, the interconnect resistance, being proportional to the
square of the scaling factor, has increased significantly.
An increase in interconnect resistance has led to an increase ininterconnect delays thus making them a dominant factor in the timinganalysis of VLSI circuits.
Therefore VLSI circuit designers aim at finding minimum cost(spanning or Steiner) trees given delay bound constraints on source-sink connections.
5.1 Introduction of MNP
For example, the problem of finding low-cost and low-transmission-delay
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 140/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 140
p , p g ymultimedia networks can be modeled as the (Diameter, Total cost,Spanning tree)-bicriteria problem:
given an undirected graph G = (V ,E ) with two weight functions c e and d e for each edge e∊E modeling construction and delay costs respectively, and abound D (on the total delay), find a minimum c -cost spanning tree such thatthe diameter of the tree under the d -costs is at most D.
It is easy to see that the notion of bicriteria optimization problems can be easilyextended to the more general multicriteria optimization problems.
The applications set the stage for the formal definition of multicriterianetwork design problems. Marathe et al. explain this concept by giving aformal definition of a bicriteria network design problem. Marathe, M. V., R. Ravi, R. Sundaram, S. S. Ravi, D. J. Rosenkrantz, and H.
B. Hunt: “Bicriteria network design problems,” J. Algorithms, vol. 28, no. 1, pp.142-171, 1998.
Marathe et al. study the complexity and approximability of a number of bicriteria network design problems. The three objectives considered: total cost
diameter
degree of the network.
5.2 Reviewing Solution Approaches for MNP
a AWA (Gen et al 1998)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 141/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 141
a. AWA (Gen et al., 1998) Gen, M. & R. Cheng: Genetic Algorithms and Engineering
Optimization, John Wiley & Sons, New York, 2000.
b. RWA (Murata et al., 1998) Gen, M. & R. Cheng: Genetic Algorithms and Engineering
Optimization, John Wiley & Sons, New York, 2000.
c. SPEA (Zitzler et al., 1999) Zitzler, E. & L. Thiele: “Multiobjective Evolutionary Algorithms: A
Comparative Case Study and the Strength Pareto Approach”, IEEE
Trans. on Evol. Comput., Vol.3, No.4, pp.257-271, 1999.
d. NSGA- (DebⅡ et al., 2000) Deb, K., A. Pratap, S. Agarwal and T. Meyarivan: “A Fast and Elitist
Multiobjective Genetic Algorithm: NSGA- ”,Ⅱ IEEE Trans. on Evol.
Comput., Vol.6, No.2, 182-197, 2002.
5.2 Reviewing Solution Approaches for MNP
a AWA (Gen & Cheng 1998)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 142/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 142
a. AWA (Gen & Cheng, 1998)
Gen & Cheng (1998) proposed an Adaptive Weight Approach (AWA)
which utilizes some useful information from the current population toreadjust weights to obtain a search pressure toward a positive ideal point.
Gen, M. & R. Cheng: Genetic Algorithms and Engineering Optimization, John Wiley & Sons,
New York, 2000.
For the examined solutions at each generation, they define two extreme points
(maximum: z +, minimum: z -)
where z k max and z k
min are the maximal and minimal values for the k th objective as
defined by the following equations:
][
][
minmin
2
min
1
maxmax
2
max
1
q
q
z z z
z z z
=
=
−
+
z
z
qk P f z
qk P f z
k k
k k
,,2,1},|)(min{
,,2,1},|)(max{
min
max
=∈=
=∈=
x x
x x
P : set of solution candidates.
5.2 Reviewing Solution Approaches for MNP
a. AWA (Gen & Cheng, 1998)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 143/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 143
a. AWA (Gen & Cheng, 1998) The weighted-sum objective function for a given chromosome x is given by the
following equation:
where w k is adaptive weight for objective k :
The equation driven above is a hyperplane defined by the following extreme points
in current solutions:
∑∑∑=== −
−=
−
−=−=
q
k k k
k k q
k k k
k k q
k
k k k z z
z f
z z
z z z z w z
1
minmax
min
1
minmax
min
1
min )()()(
x x
qk z z
wk k
k ,,2,1,1
minmax=
−=
][minminmin
2
max
1 qk z z z z
][minminmax
2
min
1 qk z z z z
][maxminmin
2
min
1 qk z z z z
][ minmaxmin
2
min
1 qk z z z z
5.2 Reviewing Solution Approaches for MNP
a. AWA (Gen & Cheng, 1998)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 144/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 144
Fig.6.7 Adaptive weights and adaptive hyperplane
+
z
1 z
2 z
min
2 z
max
1 z
max2 z
min
1 z
subspace
corresponding to
current solutions
adaptivemoving line
whole criteria space Z
positive ideal pointminimal rectangle containing
all current solutions
maximumextreme point
minimumextreme point
− z
),(max
2
min
1z z
),(min
2
max
1 z z
( & g, )
Adaptive moving line defined by the extreme points ( z 1
max, z 2
min) and ( z 1
max, z 2
min) are
shown as follows:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 145/165
5.2 Reviewing Solution Approaches for MNP
b. RWA (Murata et al., 1998)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 146/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 146
( , )
For a problem to maximize q objective functions, weighted-sum objective
is given as the follows:
Random-weight wk is calculated by the following equation:
where r j are non-negative random number between [0, 1].
Before selecting a pair of parents for crossover operation, a new set of
random weights is specified. The selection probability pi for individual i is
then defined by the following linear scaling function:
where z min is the worst fitness value in the current population.
∑=
=q
k
k k f w z 1
)( x
qk
r
r w
q
j j
k k ,,2,1,
1
==
∑=
( )∑ =−
−= popSize
j j
ii
z z z z p
1 min
min
5.2 Reviewing Solution Approaches for MNP
c SPEA (Zitzler et al 1999)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 147/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 147
c. SPEA (Zitzler et al., 1999)
Zitzler & Thiele (1999) proposed a new evolutionary approach to multicriteria
optimization, the Strength Pareto Evolutionary Algorithm (SPEA), that
combines several features of previous multiobjective EA’s in a unique manner.
Zitzler, E. & L. Thiele: “Multiobjective Evolutionary Algorithms: A Comparative Case Study and
the Strength Pareto Approach”, IEEE Trans. on Evol. Comput., Vol.3, No.4, pp.257-271, 1999.
It is characterized by:
Storing nondominated solutions externally in a second, continuously updatedpopulation.
Evaluating an individual’s fitness dependent on the number of external
nondominated points that dominate it.
Preserving population diversity using the Pareto dominance relationship.
Incorporating a clustering procedure in order to reduce the nondominated set withoutdestroying its characteristics.
5.2 Reviewing Solution Approaches for MNP
c. SPEA (Zitzler et al., 1999)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 148/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 148
The fitness assignment procedure is a two-stage process.
First, the individuals in the external nondominated set P ’ are ranked.
where si is proportional to the number of population members j∈ P for
which i ≻ j. n is the number of individuals in P that are covered by i
and N is the size of P .
1+=
N
n si
f 1
f 2
3/8
5/8
3/8
Fig. 6.9 Two scenarios for a maximization problem with two objectives.
5.2 Reviewing Solution Approaches for MNP
c. SPEA (Zitzler et al., 1999)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 149/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 149
( , )
The fitness assignment procedure is a two-stage process.
Afterwards, the individuals in the population P are evaluated.
where the fitness of an individual j∈ P is calculated by summing the
strengths of all external nondominated solutions i∈ P’ that cover j.
f 1
f 2
3/8
5/8
3/8
Fig. 6.10 Two scenarios for a maximization problem with two objectives.
),1[,1,
N f s f j
jii
i j ∈+= ∑ where
11/8
16/8
19/8
13/813/8
16/8 11/8
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 150/165
5.2 Reviewing Solution Approaches for MNP
d. NSGA- (DebⅡ et al., 2000)
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 151/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 151
( , )
The new population P t +1 is now used for selection, crossover, and mutation to
create a new population Qt +1. It is important to note that they used a binary tournament selection operator ,
but the selection criterion is based on the crowded-comparison operator ≺n.
Crowded-comparison operator is defined as follows:
where, Ri is nondomination rank and Di is crowding distance
( ) ( ) ji
D D R R R R
n
ji ji ji then
andorif )()( >=<
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 152/165
5.3 Numerical Examples
Test Problems:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 153/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 153
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25 f
10, 20
13, 20
32, 20
135, 20
631, 20
8, 20
6, 20
7, 20
7, 20
7, 20
5, 20
14, 20
4, 20
14, 20
10, 10
35, 10
3, 10
33, 10
7, 10
7, 10
3, 10
10, 10
13, 8
15, 8
11, 8
5, 8
3, 8
11, 25
8, 25
35, 25
14, 25
12, 25
34, 20
10, 15
4, 15
9, 15
11, 15
12, 15
9, 15
14, 15
5, 15
10, 6
15, 5
33, 4
4, 5
13, 4
10, 30
2, 30
3, 30
f
The first numerical example, presented by Munakata & Hashier , was
adopted. The problem comprises 25 nodes and 49 arcs. It is given asfollows:
i jcij , uij
5.3 Numerical Examples Test Problems:
The second numerical example presented by T Munakata & D J Hashier was adopted
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 154/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 154
The second numerical example, presented by T. Munakata & D.J. Hashier , was adopted.
The problem comprises 25 nodes and 56 arcs. It is given as follows:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25 f f
2, 20
10, 20
33, 20
126, 20
634, 20
3, 10
14, 8
8, 10
13, 6
32, 8
7, 5
12, 9
33, 8
4, 7
12, 10
30, 7
11, 12
12, 15
3, 5
7, 18
12, 7
4, 6
6, 12
12, 8
9, 7
8, 20
12, 6
3, 5
11, 22
6, 6
29, 11
13, 2
11, 5
30, 10
6, 8
9, 10
34, 10
15, 10
2, 9
35, 6
14, 5
7, 8
8, 10
6, 8
26, 9
12, 7
31, 15
8, 4
30, 8
11, 6
35, 9
2,15
10, 20
6, 30
7, 20
9, 15
i jcij , uij
5.3 Numerical Examples
Reference solution set S* :
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 155/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 155
Reference solution set S : The reference solution set S * of each test problem was
found using the SPEA, NSGA-2, RWA, and AWA. Each
algorithm was applied to each test problem with much
longer computation time and larger memory storage than
the other computational experiments in this study.
More specifically, we used the following parameter specifications in all the three algorithms for finding the
reference solution set of each test problem. Population size: popSize =30
Crossover probability: pC =0.70
Mutation probability: pM =0.80 Immigration probability: μ=5
Stopping conditions: Evaluation of 100000 solutions.
5.3 Numerical Examples
Reference solution set S * to test problems:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 156/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 156
p
0
5000
10000
15000
20000
0 20 40 60 80 100
flow
c o s
t
Fig. 6.11 The Reference solution set of Example 1 (|S *|=69)
5.3 Numerical Examples
Reference solution set S* to test problems:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 157/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 157
Reference solution set S to test problems:
z 1 z 2 z 1 z 2 z 1 z 2 z 1 z 2
5 260 39 2380 62 5339 82 14088
6 318 40 2447 63 5743 83 14841
8 376 41 2531 64 5936 84 15561
10 510 42 2648 65 5962 85 16198
11 590 43 2696 66 6162 86 16842
12 644 45 2866 67 6382 87 17528
13 744 46 3042 68 6714 88 1833216 808 47 3049 69 6846 89 18988
18 918 48 3151 70 6952 90 19597
20 1030 49 3322 71 7182
23 1206 50 3393 72 7364
25 1320 51 3527 73 8043
28 1496 52 3530 74 8707
30 1650 53 3932 75 9375
32 1796 54 4066 76 1005733 1875 55 4072 77 10750
34 1955 56 4364 78 11438
35 2058 57 4479 79 12123
37 2162 58 4747 80 12735
38 2262 60 5010 81 13523
Table 6.6 The Reference solution set of Example 1 (|S *|=69)
5.3 Numerical Examples
Reference solution set S * to test problems:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 158/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 158
0
5000
10000
15000
20000
0 20 40 60 80 100
p
flow
c o s
t
Fig.6.12 The Reference solution set of Example 2 (|S *|=77)
5.3 Numerical Examples
Reference solution set S * to test problems:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 159/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 159
z 1 z 2 z 1 z 2 z 1 z 2 z 1 z 2 2 52 31 1408 55 3525 75 6815
5 115 32 1488 56 3610 76 7081
7 167 33 1595 57 3669 77 7227
8 248 34 1640 58 3732 78 7308
10 275 35 1753 59 3987 79 8002
11 306 36 1818 60 4189 80 8697
13 388 37 1892 61 4458 81 9319
15 495 38 1942 62 4551 82 9997
17 608 40 2171 63 4751 83 10667
18 653 42 2319 64 4867 84 11341
20 784 43 2405 65 5069 85 11949
22 872 44 2587 66 5341 86 12623
23 953 45 2653 67 5517 87 13455
24 965 47 2701 68 5583 88 14116
25 1035 48 2897 69 5809 89 14735
26 1168 49 2974 70 5941 90 15551
27 1217 50 3077 71 6128 91 16100
28 1275 51 3106 72 6319
29 1332 52 3198 73 6487
Table 6.7 The Reference solution set of Example 2 (|S *|=77)
5.3 Numerical Examples
Performance Measures:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 160/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 160
We mainly use a performance measure based on: The number of obtained solutions |S j |
The ratio of nondominated solutions R NDS(S j )
The R NDS(S j ) measure can be written as follows:
The distance D1R
The D1Rmeasure can be written as follows:
where S* is a reference solution set for evaluation the solution set S j .
d xr is the distance between a solution x and a reference solution r .
[Ref.] Ishibuchi, H., T. Yoshida & T. Murata: “Balance Between Genetic Search and Local Search in
Memetic Algorithms for Multiobjective Permutation Flowshop Scheduling”. IEEE Trans. On Evol.
Comp., Vol. 7, No. 2, pp. 204-223, 2003.
[Ref.] Ishibuchi, H., T. Yoshida & T. Murata: “Balance Between Genetic Search and Local Search in
Memetic Algorithms for Multiobjective Permutation Flowshop Scheduling”. IEEE Trans. On Evol.
Comp., Vol. 7, No. 2, pp. 204-223, 2003.
{ }
j
j j
j NDS S
xr S r S xS S R
:*)(
∈∃∈−=
∑∈
∈=*
R }min{*
1D1
S r
jrx S xd S
( ) ( )222
2
11 )()()()( x f r f x f r f d rx −+−=
5.3 Numerical Examples
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 161/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 161
Table 6.8 Comparison with the four approaches using the |S j| measure.
Test Problems(# of nodes/ # of arcs)
|S j | CPU TimesRWA SPEA NSGA-Ⅱ AWA RWA SPEA NSGA-Ⅱ AWA
25/49 52 57 43 49 15122 17635 15693 14170
25/56 43 44 55 43 11918 16684 15981 14961
Table 6.9 Comparison with the four approaches using the RNDS (S j) measure.
Test Problems
(# of nodes/ # of arcs)
RNDS (S j ) CPU Times
RWA SPEA NSGA-Ⅱ AWA RWA SPEA NSGA-Ⅱ AWA
25/49 0.57 0.54 0.39 0.61 15122 17635 15693 14170
25/56 0.41 0.34 0.36 0.53 11918 16684 15981 14961
Test Problems
(# of nodes/ # of arcs)
D1R measure CPU Times
RWA SPEA NSGA-Ⅱ AWA RWA SPEA NSGA-Ⅱ AWA
25/49 191.21 315.61 228.65 143.58 15122 17635 15693 14170
25/56 203.96 224.40 185.89 141.43 11918 16684 15981 14961
Table 6.10 Comparison with the four approaches using the D1R measure.
5.3 Numerical Examples
Different Parameter Settings:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 162/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 162
Comparison with different approaches using the stopping conditions: under the
same computation time: 10,000 ms.
Table 6.11 Comparison with the four approaches using the |S j| measure.
Test Problems
(# of nodes/ # of arcs)
|S j |
AWA RWA SPEA NSGA-Ⅱ
25/49 49 50 53 50
25/56 52 50 40 34
Table 6.12 Comparison with the four approaches using the RNDS (S j) measure.
Test Problems
(# of nodes/ # of arcs)
RNDS (S j )
AWA RWA SPEA NSGA-Ⅱ
25/49 0.57 0.44 0.56 0.48
25/56 0.51 0.32 0.60 0.41
Test Problems
(# of nodes/ # of arcs)
D1R measure
AWA RWA SPEA NSGA-Ⅱ
25/49 191.17 203.72 222.28 239.99
25/56 147.07 219.59 279.60 433.43
Table 6.13 Comparison with the four approaches using the D1R measure.
5.3 Numerical Examples
Different Parameter Settings:
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 163/165
Soft Computing Lab. WASEDA UNIVERSITY , IPS 163
Comparison with different approaches using the stopping conditions: under the
same computation time: 10,000 ms.
Table 6.11 Comparison with the four approaches using the |S j | measure.
Test Problems
(# of nodes/ # of arcs)
|S j |
RWA SPEA NSGA-Ⅱ AWA
25/49 50 53 50 49
25/56 50 40 34 52
Table 6.12 Comparison with the four approaches using the RNDS (S j ) measure.
Test Problems
(# of nodes/ # of arcs)
RNDS (S j )
RWA SPEA NSGA-Ⅱ AWA
25/49 0.44 0.56 0.48 0.57
25/56 0.32 0.60 0.41 0.51
Test Problems
(# of nodes/ # of arcs)
D1R measure
RWA SPEA NSGA-Ⅱ AWA
25/49 203.72 222.28 239.99 191.17
25/56 219.59 279.60 433.43 147.07
Table 6.13 Comparison with the four approaches using the D1R measure.
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 164/165
Conclusion In this study, we presented a GA approach used a priority-based chromosome
for solving the network design problems
8/6/2019 ssp, mst, msf
http://slidepdf.com/reader/full/ssp-mst-msf 165/165
for solving the network design problems.
It is easy to verify that any permutation of the encoding corresponds to the
paths.
So that most existing genetic operators can easily be applied to the
encoding.
Also, any path has a corresponding encoding.
Therefore, any point in solution space is accessible for genetic search.
For solving the MXF/MCF, and Multi-criteria Network Design Problem, we alsocombines an adaptive evaluation function based on the AWA.
The fitness values of all individuals are calculated according to this
adaptive evaluation function.
In each generation, the set of Pareto solutions is updated by deleting all
dominated solutions and adding all newly generated Pareto solutions. Computer simulations show the several numerical experiments by using