Automated Information Retrieval Model Using FP Growth Based Fuzzy Particle Swarm Optimization

International Journal of Computer Science & Information Technology (IJCSIT) Vol 7, No 1, February 2015
DOI:10.5121/ijcsit.2015.7106 59
IMPROVING INITIAL GENERATIONS IN PSO
ALGORITHM FOR TRANSPORTATION NETWORK
DESIGN PROBLEM
Navid Afkar and Abbas Babazadeh
Department of Civil Engineering, University of Tehran, Tehran, Iran
ABSTRACT
Transportation Network Design Problem (TNDP) aims to select the best project sets among a number of
new projects. Recently, metaheuristic methods are applied to solve TNDP in the sense of finding better
solutions sooner. PSO as a metaheuristic method is based on stochastic optimization and is a parallel
revolutionary computation technique. The PSO system initializes with a number of random solutions and
seeks for optimal solution by improving generations. This paper studies the behavior of PSO on account of
improving initial generation and fitness value domain to find better solutions in comparison with previous
attempts.
KEYWORDS
Transportation; Network Design; optimization; Particle Swarm; roulette cycle; initial value
1. INTRODUCTION
In transportation planning, Transportation Network Design Problem (TNDP) is a substantial area
in which specific objectives are minimized through selection among a given set of projects under
constraints (1). Solving TNDP requires too much time. Various approaches have been taken to
solve TNDP (1,2,3). One of the typical methods to solve TNDP are Meta-Heuristic algorithms
(such as Genetic Algorithm). Particle Swarm optimization (PSO) is a Meta-Heuristic algorithm
that has shown good performance to solve TNDP.
In this paper, particle swarm optimization (PSO) algorithm developed with some variations or
added methods is presented to solve the TNDP (4,5). The results of each method are compared
together and with Original PSO method. The remainder of the paper is organized as follows. The
next section is devoted to define the TNDP mathematically. In the following sections, the PSO is
described in details, and then applied to the TNDP. After that Binary Sorting, Binary Count, and
Roulette methods are defined and used to enhance the efficiency of PSO for solving the TNDP on
the Sioux Falls network. The results are obtained by a computer program in VISUAL BASIC 6.0
on a laptop with Intel core 2 due 2.4 GHz processor. In this program, each algorithm is terminated
after a fixed number of 1000 iterations. Due to the stochastic nature of PSO, the algorithms have
been solved 50 times and the results are based on the average values of the 50 runs.
Computational results and figures are reported in the final section.

60
2. TNDP
Let ),( AVG  be a graph representing a transportation network with node set V and arc set A,
and define }:),{( srVVsrP  as the set of origin-destination (OD) pairs. For each OD
pair Psr ),( , there is a nonnegative flow rate (travel demand) from r to s, denoted by rsd . In
order to simplify the presentation, suppose that G is strongly connected, that is each node j can
be reached from every other node i by following a directed path in G , and let rsK be the non-
empty set of paths from the origin r to the destination s.
Define Ā (Ā ≠ ) as the set of project arcs, and let the decision vector be y = ( ) ∈Ā with
being the binary project decision variable, taking values 0 or 1 depending on rejection or
acceptance of any project ∈ Ā . For a given vector y, define the decision network G( ) =
( , ( )) with A(y) = A∪ { ∈ Ā ∶ = 1} as the set of arcs followed by decision y, and for
each (r, s) ∈ denote by ( ) the set of paths joining r to s in G( ). For each path k ∈ ( )
let be the flow of path k from origin r to destination s. Moreover, let equals 1 if arc a
∈ ( ) lies on path k, and 0 otherwise.
Assume further that each arc a ∈ ∪ Ā has a node creasing and continuously differentiable
travel time function ( ):[0, ∞) → [0, ∞) with being the flow rate assigned to arc a. Then,
letting be the construction cost of project arc a ∈ Ā , and considering the total construction
cost being limited to the level of Budget B, the TNDP can be illustrated with upper level problem,
ULP:
[ULP] T(y) = ∑ ( ∈ ( ) )
s.t. ∑ ( ) ∈ Ā
=0 or 1 ∀a ∈ Ā
X(y) is a solution of [LLP(y)]
Where x(y) = ( ) ∈ ( ) is the user equilibrium flow in the decision network G(y), given as
the solution of the lower level (traffic assignment) problem, LLP(y), for given y:
[LLP(y)] Min ∑ ∫ ( )∈ ( )
s.t. ∑ ∈ ( ) = ∀( , ) ∈
≥ 0 ∀ ∈ ( ), ∀ ( , ) ∈
= ∑ ∑ ∈( , )∈ ∀ ∈ ( )
This is a well-known bi-level programming problem, where the [ULP] seeks a decision vector
y for minimizing the total travel time T(y) of the (assigned) traveler, and the [LLP(y)] is the traffic
assignment model which estimates the traveler flows, given the decision y (6,7,8).
3. PARTICLE SWARM OPTIMIZATION
Particle Swarm Optimization (PSO) is a meta-heuristic optimization approach which has been
widely applied to various problems (4). PSO technique that was developed by Kennedy and

61
Eberhart is originated from the behavior of birds ’flocks in which individuals convey information
between themselves and the leader in order to seek the best direction to food (9, 10).
In a problem space, each particle has a position and a velocity and it moves in the search space
with the velocity according to its own previous best position and the group’s previous best
position. The dimension of the search space can be any positive number. Considering D as the
dimension of the search space, the ith
particle’s position and velocity are represented as
  1,...,i ij j D
P p

 and   1,...,i ij j D
V v

 respectively. Each particle maintains its own best position so
far achieved as  * *
1,...,i ij j D
P p

 and the global best position so far recorded by the population as
 * *
1,...,g gj j D
P p

 .
During the iteration time t, the velocity of the jth
dimension of each particle i is updated by:
* *
1 1 2 2( 1) ( ) ( ( ) ( )) ( ( ) ( ))ij ij ij ij gj ijv t wv t c r p t p t c r p t p t      Where w is called as the
inertia weight, 1c and 2c are constant values and 1r , 2r are random numbers in the interval 0,1 .
The current position of each particle is then defined by the sum of its current velocity and its
previous position (11,12,13).
( 1) ( ) ( 1)ij ij ijp t p t v t   
In order to avoid the particles from moving out of the search space, the maximum velocity during
the iterations is restricted by maxv . As proposed by Hong Zhang, et al 2005, the maximum velocity
( maxv ) is set to maxx . This results in moving more effectively in the search space and accordingly
better algorithm performance.
4. ADAPTING THE PSO TO THE TNDP
Employing the PSO for solving TNDP needs some modifications to the algorithm given in the
previous section. First, the PSO is basically developed for continuous optimization problems(14).
This is while the TNDP is formulated as a combinatorial optimization problem in terms of
variables y denoted as A -bit binary strings. To adapt the algorithm for this combinatorial
nature, one may provide some mapping from the one-dimensional real-valued space to the A -
dimensional binary space. This is done here by transforming each real number ip to its nearest
integer in 0 , 2 1
A 
  
, and then transforming the resulting integer in to the base-2 number
system as an A -bit binary code. To facilitate the presentation, the latter transformation is
illustrated by the function  ( ) : 0 , 2 1 0 ,1 .
A A
iy p Z   
  
The PSO must also be adapted for budget constraint embedded in the [ULP](15).

62
5. PSO ALGORITHM
Step 1. Initialization
Select the particle swarm size n, the parameters 1c and 2c , the value of the inertia weight w, and
the maximum velocity maxv .
For i 1 to n do: initialize the decision variable ip so that Bpyc iaaAa
 
)( ; set
and 0iv .
Set ))(..,.),((minarg 1 ng pfpfp 
. Set the iteration counter 0t .
Step 2. Updating each particle's position and velocity
For i 1 to n do: generate random numbers 1r and 2r in [0, 1]; update
)()( 2211 igiiii pprcpprcwvv  
; clamp in iv between the range ],[ maxmax vv as
)|,min(|)sign( maxvvvv iii  ; update iii vpp  ; transform ip to its nearest integer in
12,0[ ||
A
].
Step 3. Calculating each particle's fitness value
For i 1 to n do: set )( ipyy  ; if Byc aaAa
 
then set Mpf i )( (large fitness value);
else, solve the user equilibrium problem [LLP ( y )] to compute )(yT , and set )()( yTpf i  .
Step 4. Updating local bests and global best
For i 1 to n do: update ))(),((minarg iii pfpfp 
 .
Update ))(..,.),(),((minarg 1 ngg pfpfpfp 
 .
Step 5. End criterion.
Set 1 tt . If end criterion is not met, go to Step 2. Otherwise, )( 
 gpyy is the best solution
found so far with the objective function value )()( 
 gpfyT Collect the necessary information
and stop.
6. SIOUX FALLS NETWORK
The Sioux Falls network has 24 nodes and 76 arcs, as shown in Fig. 1. The parameters of the
travel time function
4
)( aaaaa xxt   for each arc a, and the OD (origin/destination)
demands are basically those given in Poorzahedy and Turnquist (1982), and LeBlanc (1975), and
are eliminated here for brevity(16).
There are 10 pairs of project arcs )10|(| A , of which 5 projects are improvement on existing arcs,
and 5 are new arcs. The construction costs of the projects 1-10 are, respectively, 625, 650, 850,
1000, 1200, 1500, 1650, 1800, 1950, and 2100 units of money (Poorzahedy and Abulghasemi
2005). Considering 10 projects, there are )1024(210
 alternative networks. A complete
enumeration was used to compute the optimal solution of the TNDP for any given budget level
for checking purposes (Poorzahedy and Abulghasemi 2005; Poorzahedy and Rouhani 2007).
ii pp 

63
7. BINARY SORTING (BS)
Binary Sorting is a type of mapping the objective function value domain from
:[0 ,2 1]n
ix Z  to :[0 ,2 1]n
mp Z  . This results in a more effective sorting on the
function domain and puts the decision variables with the same number of projects beside each
other, therefore the PSO algorithm searches with more intelligence toward the optimum solution.
The algorithm is expressed below:
Let iB be the binary value of :[0 ,2 1]n
ix Z  in base-2 number system then count the iB
number of digits and put them into iC .
m=1
for d=0 to n
for i=0 to 2n
-1
if > −1 then
if iC =d then
m ip x
1ix  
m = m+1
End if
End if
Next i
Next d
8. BINARY COUNT (BC)
Binary Count is an initialization strategy which is defined for this specific problem. The idea
comes from choosing more particles which are near to the budget level. The algorithm is outlined
below.
For d=0 to n
For i=0 to 2n
-1
If Ci=d then count (d) =count (d) + 1
Next i
Next d
Assume b/c
all=int(b/c * n)+1
For i=0 to all
numb(i)=i / ∑ * n
sum numb= numb(i)+sumnumb
Next i
m=1
for i=1 to all
for d=1 to numb(i)
Select a particle that has i binary digit
Next d
Next i
Do while sum numb(i) < n

64
Select a particle that has at most all binary digits
Loop
9. ROULETTE WHEEL SELECTION (ALGORITHM)
It is the most common selection strategy. It will assign to each individual a selection probability
that is proportional to its relative fitness. Let be the fitness of the individual in the population
P. Its probability to be selected is = / (∑ ).Suppose a pie graph where each individual is
assigned a space on the graph that is proportional to its fitness. An outer roulette wheel is placed
around the pie. The selection of μ individuals is performed by μ independent spins of the roulette
wheel. Each spin will select a single individual. Better individuals have more space, and then
more chance to be chosen. Moreover, when all individuals are equally fit, this selection strategy
does not introduce a sufficient pressure to select the best individuals. This method is applied on
the initialization step of the PSO algorithm to improve its initial generation and the combined
algorithm is named as Roulette in the following discussions.
10. COMPUTATIONAL RESULTS
In this study, the algorithms comprised of the above strategies are compared from various
perspectives. First, Average Objective Function Value (OFV) related to the initial generation of
each algorithm is shown in figure 2. According to this figure, Roulette and Roulette-BS (Binary
Sorting) generate better initial particles than the other methods.
In figure 3, BC method shows a good convergence performance concerning its decreasing
behavior despite of the relative large quantity of the initial average OFV and after 70 iterations
average OFV of BC method gets very close to average OFV of the Roulette-BS. Still, it is
obvious that the Roulette-BS method shows the best performance based on initialization and
convergence capability and its graph stands under the others in figure 3.
To perceive the effectiveness of Binary Sorting, Random and Random-BS methods are compared
with each other. The Random-BS method has more decreasing behavior than the Random method
and shows more convergence to the optimum solutions in the last iterations. So we can figure out
that the BS method makes an improvement on the PSO algorithm which is adapted to the TNDP.
Figure 4 depicts the frequency of finding the optimal solution in iterations for each method. It is
clear that each algorithm graph that stands upper than the other graphs is more powerful to find
the optimum solution. Comparison of these methods proves that the Roulette and Random
methods have less capability than the other three methods which have used binary sorting to find
the optimum solution.
The average Number of Traffic Assignment Problem Solved (NTAPS) of each method is shown
in figure 5. All of the graphs are showing decreasing behavior. This fact proves that all methods
are trying to decrease the NTAPS. As the results show, Roulette and Random graphs stand upper
than others after 50 iterations.
Figure 6 displays initialization time and PSO algorithm time usage in each method. Initialization
time of BC method is shorter than the other methods. It’s obvious that initialization time for BC

65
method is shorter than Roulette and Roulette-BS methods, but the reason why this time is shorter
than Random method’s time is related to unsuccessful tries of generating particles with higher
budget than the problem’s budget constraint in Random method. As a result, we can figure out
that in TNDP problems with large scale of particles, the BC method shows better performance
and can find the optimum solution sooner.
REFERENCES
1. Dantzig G D, Harvey R P, Lansdowne Z F, Robinson D W and Maier S F (1979). Formulating and
solving the network design problem by decomposition. Transport Res B-Meth 13:5–17.
2. Hoang H H (1982). Topological optimization of networks: A nonlinear mixed integer model
employing generalized benders decomposition. IEEE T Automat Contr 27: 164–169.
3. LeBlanc L J (1975). An algorithm for discrete network design problem. Transport Sci 9: 183–199.
4. Eberhart R C and Kennedy J (1995). A new optimizer using particle swarm theory. Proceedings of the
Sixth International Symposium on Micro Machine and Human Science, Nagoya, Japan, pp 39-43.
Piscataway,NJ: IEEE Service Center.
5. Abraham A, Guo H and Lio H (2006). Swarm intelligence: Foundations, perspectives and
Applications. In: Nedjah N and M.Mourelle L (eds). Swarm Intelligent Systems. Springer:
Netherlands, pp 18-25.
6. Chen M and Sul Alfa A (1991). A Network design algorithm using a stochastic incremental traffic
assignment approach. Transport Sci 25: 215–224.
7. Magnanti T L and Wong R T (1984). Network design and transportation planning: models and
algorithms. Transport Sci 18 : 1–55.
8. Haghani, A.E., and Daskin, M.S. (1983). “Network design application of an extraction algorithm for
network aggregation.” Transportation Research Record, 944, 37-46.
9. Holmberg, K., and Hellstrand, J. (1998). “Solving the uncapacitated network design problem by a
lagrangean heuristic and branch-and-bound.” Operations Research, 46 (2), 247–259.
10. Yisu, J., Knowles, J., Hongmei, L., Yizeng, L., and Kell D.B. (2008). “The landscape adaptive
particle swarm optimizer.” Applied Soft Computing, 8 (1), 295-304.
11. Langerudi, Mehran Fasihozaman; Javanmardi, Mahmoud; Mohammadian, Abolfazl (Kouros); Sriraj,
PS; "Choice Set Imputation”, Transportation Research Record: Journal of the Transportation
Research Board, 2429, 1, 79-89, 2014, Transportation Research Board of the National Academies.
12. Langerudi, Mehran Fasihozaman; Abolfazl, Mohammadian; Sriraj, PS; "Health and Transportation:
Small Scale Area Association”, Journal of Transport & Health, 2014, Elsevier,
doi:10.1016/j.jth.2014.08.005
13. Fasihozaman Langerudi, Mehran; Hossein Rashidi, Taha; Mohammadian, Abolfazl; "Investigating
the Transferability of Individual Trip Rates: Decision Tree Approach", Transportation Research
Board 92nd Annual Meeting, 13-0218, 2013. http://trid.trb.org/view.aspx?id=1240410H
14. H. Miar Naimi, M. Salarian, “A Fast fractal Image Compression Algorithm Using Predefined Values
for Contrast Sacaling”, Proceedings of the World Congress on Engineering and Computer Science
USA, October-2007.
15. M Salarian, H Hassanpour, A new fast no search fractal image compression in DCT domain, Machine
Vision, 2007. ICMV 2007. International Conference on, 62-66.
16. M. Javanmardi, M. Fasihozaman, A. Talebpour, A. Mohammadian “Integrated Demand and Supply
Model: Networkwide Validation” Presented in the Transportation Research Board 93th Annual
Meeting, 2014
17. M Salarian, E Nadernejad and H. M. Naimi, A new modified fast fractal image compression
algorithm, Imaging Science Journal, vol. 61, Feb. 2013, pp. 219-231, doi: 10.1179/
1743131X11Y.0000000027.

66
Figure 1. The Sioux Falls Network
Figure 2. Average Objective Function Value for first 10 Particles

67
Figure 3. Average Objective Function Value
Figure 4. Frequency of Finding

68
Figure 5. Average Number of Traffic Assignment Problem Solved
Figure 6. Initialization time and PSO algorithm time

Automated Information Retrieval Model Using FP Growth Based Fuzzy Particle Swarm Optimization

More Related Content

What's hot (17)

Similar to Automated Information Retrieval Model Using FP Growth Based Fuzzy Particle Swarm Optimization (20)

More from AIRCC Publishing Corporation (20)

Recently uploaded (20)

Automated Information Retrieval Model Using FP Growth Based Fuzzy Particle Swarm Optimization