Professional Documents
Culture Documents
Hybrid Multiobjective Evolutionary Algorithm
Hybrid Multiobjective Evolutionary Algorithm
Hybrid Multiobjective
Evolutionary Algorithm Solving
Vehicle Routing Problem
Lecturer: Prof. Martin Henz
CS 6211 Special Topics in Computer Science II
Recombination
Population
Mutation
Offspring
Replacement
EA Application In:
Bio- computing
Evolvable hardware
Game playing
Job-shop scheduling
Management sciences
Non-linear filtering
MOEA
Multiobjective optimization problem (MOP)
A problem to formulate a design in which there are
several criteria or design objectives
MOEA
Multiobjective evolutionary algorithm
EA that can solve MOP
MOEA
Minimize [f1(x), f2(x), …, fk(x)]
Subject to the m inequality constraints
gi ( x ) 0 i 1, 2,..., m
And subject to the p equality constraints
hi ( x) 0 i 1, 2,..., p
k is the number of objectives
Conventional MO handling
Weighting method
Constraints method
Penalty method
Disadvantages
Priori
Trial and error (weights)
Pareto Concept
Two objectives
minimization.
Point A and point B
are dominating
points.
Point C is
dominated.
Curve from A to B
is Pareto front.
MOEA characteristics
Random vs. Deterministic
Population vs. Single best solution
Creating new solutions through crossover
Creating new solution through mutation
Selecting solutions via “Survival of the fittest”
MOEA algorithms
MOEA in literature
VEGA, David Schaffer (1985)
MOGA, Fonseca and Fleming (1993)
NPGA, Horn et al. (1994)
NSGA, Srinivas and Deb (1994)
PAES, Knowled and Corne (2000)
SPEA, Zitler (2001)
Research
6
7 4
3
10
Depot
8
1
9
12
11
R1 R2
VRPTW: Methods
Graph theory
Minimal spanning tree + shortest path
Saving method
Branch and bound
Simulated annealing
Tabu search
Ant colony
VRPTW: Survey
Solomon (1987)
Golden and Assad (1988)
Desroiser et al. (1995)
Laporte et al. (2000)
Kilby et al. (2000)
Toth and Vigo (2002)
Why MOEA in VRPTW ?
Optimize for a set of solutions
Minimum dependence to priori
Global search ability
Single objective vs. Multiobjective
Better solution !
Concerns in MOEA Design
Representation of chromosome
Creative or innovative evolutionary operators
Performance metrics to compare
Feasibility handling policy
Improve existing elitism reservation
HMOEA components
Variable length representation
Route exchange cross over
Multi mode mutation
Pareto ranking
Tournament selection
Local exploitation
HMOEA flow
Start
Pre-processing
Build initial
population
Route-exchange
Crossover
Elitis m
Multimode
mutation
HMOEA Representation
Chromosome A Chromosome encodes
a complete routing
0 0 0 solution
2 1 6
5 3 10
7 4 0 A route (vehicle)
contains sequence of
0 8 customers
0
Route-Exchange Xover
R1
R1
R2
R2
Selected routes
On certain criteria
Multimode mutation
PM - Mutation rate
Merge_Routes Split_Route
Local exploitation
Intra_Route
Generate two different numbers based on the size of routing sequence of
both vehicles.
Pick two routes randomly and swaps two nodes from each route.
Lambda interchange
Scan through nodes in route A and moves the node into route B.
Repeat until a pre-defined number of nodes are shifted or the the last
node of route A is met.
Shortest path first
Rearrange the order of nodes in a particular route.
Based on its distance from the previous node.
Feasibility must be ensured
Pareto Ranking
Assigns the same smallest rank (highest fitness)
for all non-dominated individuals
The individuals are inversely ranked according
to how many individuals in the population
dominating them, -
Smaller number of vehicles but equal cost of routing
Smaller routing cost but equal of number of vehicles
Smaller routing cost and smaller number of vehicles
Tournament Selection
Tournament selection.
A tournament size of 2.
All individuals in the population are randomly
grouped into pairs.
The individual with a lower rank (higher
fitness) will be selected for reproduction.
Discussion (1)
Better population distribution
Discussion (2)
Comparison to best known
Rochat Chiang Potvin Potvin and Taillard Chiang Homberger Tan Tan
Problem and and Russell, et al., Bengio, et al., and and et al., et al., HMOEA
Class Taillard, (1996) (1996) (1996) (1996) Russell, Gehring, 2001d 2001e
(1995) (1997) (1999)
C1 10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00
828.45 909.80 861.00 838.00 828.45 828.38 828.38 851.96 841.96 827.00
C2 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.20 3.00 3.00
590.32 684.10 602.50 589.90 590.30 591.42 589.86 620.12 611.2 590.00
R1 12.58 12.50 12.60 12.60 12.25 12.17 11.92 13.20 12.91 12.92
1197.42 1308.82 1294.70 1296.83 1216.70 1204.19 1228.06 1220.0 1205.0 1187.0
R2 3.09 2.91 3.10 3.00 3.00 2.73 2.73 4.40 5.00 3.51
954.36 1166.42 1185.90 1117.70 995.38 986.32 969.95 985.69 929.6 951
RC1 12.38 12.38 12.60 12.10 11.88 11.88 11.63 13.30 12.60 12.74
1369.48 1473.90 1465.00 1446.20 1367.51 1397.44 1392.57 1366.62 1392.3 1355
RC2 3.62 3.38 3.40 3.40 3.38 3.25 3.25 5.20 5.80 4.25
1139.79 1401.50 1476.10 1360.60 1165.62 1229.54 1144.43 1108.50 1080.10 1067.00
All 427 422 427 422 416 411 406 470 471 441
57120 65201 64679 62572 57993 58502 57876 57903 56931 56262
Discussion (3)
Routing solution
For RC2-07
What’s Next
5 6
12
4
8
11
10
9
Comparison to existing models
VSP (2000)
TTRP (2002)
SDVRP (1998)
PACVRP (1981)
HVRP (1999)
Job models (1)
Free Trade Customer Depot
Zone in Warehouse
Port
Loaded Trip (first leg) Empty Trip (second leg)
Empty Trip
Depot
Tasks Type (1)
Importation job
population
4600
generations 4500
Overall
4450
improvement
4400 10
13
16
19
22
25
28
31
34
40
46
49
1
37
43
Iteration
Discussion (2)
Average truck number for 100_3_4
50
45
Average truck number Convergence test
40 for 2nd objective
35
Average truck
Truck number
30
25
number
20 Reduced over
15
10
generations
1
10
13
16
19
22
25
28
31
34
37
40
43
46
49
Generation
Discussion (3)
Truck vs. Cost : Pareto-100-1-2
5100
First
Int 1
5000
Int 2
Final
4900
Pareto dominance
4800
diagram
4700
Cost of routing
Pseudo Pareto
4600
front
4500
Improvement
4400
observed
4300
4200
4100
0 5 10 15 20 25 30 35 40
Number of truck
Discussion (4)
Comparison of Comparison of Outsourced Jobs
trailers effect 30
1 or 2
25 2 or 3
3 or 4
15
10
0
100 112 120 132
Test cases category
Conclusion to TTVRP
TTVRP maneuvers routing of containers within
specified time and resource allocated.
Interesting effect of trailer: cost, truck number
required and utilization.
Performance shows dependence on distribution
pattern of trailer resource.
Summary
Summary
VRPTW solved using HMOEA
Involves the optimization of routes for multiple
vehicles so as to meet all given constraints and to
minimize the objectives of travel distance and number
of vehicles
Lower routing cost
Better population distribution
Good convergence trace
Summary (2)
TTVRP inherits common properties of a
industry model VCPS
Trailer resource in VRP
Solved using HMOEA
Useful information derived
Future Work
Improve algorithm by distributed
computing technology: to reduce
computation time
Distributed MOEA (1)
Why distributed ?
Low cost but high performance in computation
Distribute MOEA (2)
No free lunch
Distributed algorithm MOEA design
Research on partitioning algorithm
Reduce communication overhead
Achieve similar or better performance
Involve heterogeneous system properties
References
Evolutionary optimization
Ruhul Sarker, Masoud Mohammadian, Xin Yao, Boston : Kluwer academic
publishers, c2002.