Upload
orangekaer2009
View
232
Download
0
Embed Size (px)
Citation preview
8/6/2019 03 Operations Research
1/66
The role of Operations
ResearchSolving Decisional Models for EnvironmentalManagement
8/6/2019 03 Operations Research
2/66
The search process
x(t+1)=f(x(t),u(t))
Control Variables
yes
Modification of
control variable
u(t)
Time series
Maps
Simulation Model
Simulation Results
J=J(x,u)
Performance Criterion
J
max?
Search algorithm
StopAdapted from Seppelt, 2003
8/6/2019 03 Operations Research
3/66
What is search
To search means to explore possible valuesfor the control variable u, to obtain thevalue of the state x caused by u, and toevaluate the performance J
8/6/2019 03 Operations Research
4/66
The policy
The policy defines the value of u to bechosen in front of the value of x at eachtime step t and in each spatial location z
P {Mt() ; t= 0,1, . . .
8/6/2019 03 Operations Research
5/66
Finding the policy
The problem is hard effects are uncertain
decisions are dynamic and recursive
8/6/2019 03 Operations Research
6/66
8/6/2019 03 Operations Research
7/66
Scenario analysis
Scenario definition requires knowledge onthe modelled system, which might not be
available, due to complexity (Jakeman andLetcher, 2003)
In some (most) cases Scenarios are toolimited
Too much weight on past experience
8/6/2019 03 Operations Research
8/66
Optimisation
Given the goal, the computer automaticallygenerates scenarios, by combination ofparameters, controls, exogenous inputs
A numerical optimisation algorithm is incharge of evaluating the performance and
directing exploration in the most promisingdirection
8/6/2019 03 Operations Research
9/66
Simulation explained
8/6/2019 03 Operations Research
10/66
The simulation model
It is a constraint for the optimisationproblem all model variables may vary in space
boundary and initial conditions are given
xt+!t(z) = M!t(xt(z),ut(z),"(z),z)
8/6/2019 03 Operations Research
11/66
The performance
criterion
state x and control u depend on time andlocation
the performance criterion maps a trajectoryfrom state and policy space to a real
number
J(xt(z),ut(z))
8/6/2019 03 Operations Research
12/66
The problem
xt+!t(z) = M!t(xt(z),ut(z),"(z),z)
J
(xt
(z
),ut
(z
))Performance criterion:
Model:
Problem: find u so that
x(t) = M!t(x,u,")
J(x,u) J(x,u)
for all t in [0, T] and u in U
8/6/2019 03 Operations Research
13/66
Solving the problem
Exact methods Approximation algorithms
Heuristic and metaheuristic algorithms
8/6/2019 03 Operations Research
14/66
8/6/2019 03 Operations Research
15/66
Exact methods
The algorithm depends on the problemformulation
Linear programming, simplex Integer programming, branch & bound Dynamic programming
8/6/2019 03 Operations Research
16/66
Approximation
algorithms Find a sub-optimal solution
It is possible to say how far from the
optimum (e.g. 95%)
Formally proven
Often is very specific to the problem athand
E.g: neuro-dynamic programming
8/6/2019 03 Operations Research
17/66
Heuristics
Apply rules-of-thumb to the solution of theproblem
No guarantee to find an optimum No guarantee on the quality of the solution Yet, they are fast and generic
8/6/2019 03 Operations Research
18/66
Metaheuristics
set of concepts that can be used to defineheuristic methods that can be applied to awide set of different problems.
general algorithmic frameworks which canbe applied to different optimization
problems with relatively few modificationsto make them adapted to a specific problem
8/6/2019 03 Operations Research
19/66
Metaheuristics are...
Simulated annealing
Evolutionary computation Tabu search
Ant Colony Systems
8/6/2019 03 Operations Research
20/66
Ant Colony Systems
8/6/2019 03 Operations Research
21/66
Ants
Leaf cutters,fungi growers
Breeding otherInsects
Weaver ants
I i l
8/6/2019 03 Operations Research
22/66
Insects, socialinsects, and ants
1018 living insects (estimate)
~2% of all insects are social
Examples of social insects:
All ants
All termites
Some bees
Some wasps
50% of all social insects are ants
Average body weight of an ant: 1 - 5 mg
Total mass of ants ~ total mass of humans
Ants have colonized earth for 100 million yrs; Homosapiens sapiens for 50000 years
8/6/2019 03 Operations Research
23/66
Ant colony society
Size of a colony:from a few (30) to millions of worker ants
Labour division:
Queen: reproduction
Soldiers: defense
Specialised workers: food gathering, offspring tending,nest grooming, nest building and maintenance
8/6/2019 03 Operations Research
24/66
How do ants
coordinate theiractivities?
Basic principle: stigmergy stigmergy is a kind of indirect
communication, used by social insects to
coordinate their activities
8/6/2019 03 Operations Research
25/66
StigmergyStimulation of workingants due to the results
they have reachedGrass P. P., 1959
8/6/2019 03 Operations Research
26/66
8/6/2019 03 Operations Research
27/66
Artificial Stigmergy
Indirect communication by changes on theenvironment accessible only locally tocommunicating agents
Features of artificial stigmergy : Indirect communication Local accessibility
(Dorigo and Di Caro, 1999)
8/6/2019 03 Operations Research
28/66
The bridge experiment
Goss et al., 1989, Deneubourg et al., 1990
8/6/2019 03 Operations Research
29/66
Pheromone trail
Ants and termites follow pheromone trails
8/6/2019 03 Operations Research
30/66
Asymmetric bridge
experiment
Goss et al., 1989
8/6/2019 03 Operations Research
31/66
Ant System for TSP
Pheromonetrail deposition
Probabilistic rule tochoose the path
Travelling Salesman Problem
memory
Dorigo, Maniezzo, Colorni, 1991Gambardella & Dorigo, 1995
Ph i ti
8/6/2019 03 Operations Research
32/66
Pheromone, euristicsand memory to visit the
next citMemory of
visited cities
tij ;hiji
k
j
tik ;hik
tir;hir
r
pijk
t( )= f ij t( ),ij t( )( )
r
8/6/2019 03 Operations Research
33/66
Pheromone trail
depositionij
kt+1( ) 1 ( ) ij
kt( ) + ij
kt( )
ijk
t( ) = quality kwhere (i,j) are the links visited by ant k, and
where qualityk
is inversely proportional to the lengthof the solution found by ant k
8/6/2019 03 Operations Research
34/66
8/6/2019 03 Operations Research
35/66
3 main components: TIME: a short route
gets pheromone more quickly
QUALITY: a short routegets more pheromone
COMBINATORICS: a short pathgest pheromone more frequentlysince it (usually) has a smaller numberof decision points
Why does it works?
i
j
tijd
origin
destination
d
Some results on the
8/6/2019 03 Operations Research
36/66
Some results on theTravelling Salesman
Problem
A i h i i
8/6/2019 03 Operations Research
37/66
A constructive heuristics
with local search The best strategy to approximate thesolution of a combinatorial optimisation
problem is to couple a constructive heuristics and local search
It is hard to find the best coupling: Ant System (and similar algorithms)
appear to have found it
8/6/2019 03 Operations Research
38/66
The ACO
Metaheuristics Ant System has been extended, to be appliedto any problem formulated as a shortest pathproblem
The extension has been named Optimisationmetaheuristics based on Ant Colonies (ACO)
The main application:
NP complex combinatorial optimisationproblems
8/6/2019 03 Operations Research
39/66
The ACO-metaheuristic
procedureprocedure ACO-metaheuristic()
while (not-termination-criterion)
planning subroutines
generate-and-manage-ants()
evaporate-pheromone()
execute-daemon-actions() {opzionale}
end planning subroutinesend while
end procedure
E.g. local search
8/6/2019 03 Operations Research
40/66
ACO: Applications
Sequential ordering in a production line
Vehicle Routing of trucks goodsdistributions Job-shop scheduling
Project scheduling Water distribution problems
8/6/2019 03 Operations Research
41/66
Genetic Algorithms
8/6/2019 03 Operations Research
42/66
Evolutionary computing
EC (Evolutionary computing) = GA (Genetic Algorithms - Holland, 1975) ES (Evolution Strategies - Rechenberg
1973)
EP (Evolutionary Programming - Fogel,
Owens, Walsh, 1966)
GP (Genetic Programming - Koza, 1992)
8/6/2019 03 Operations Research
43/66
Biological basis
Evolution operates on chromosomes, whichencode the structure of a living being
Natural selection favours reproduction ofthe most efficient chromosomes
Mutations (new chromosomes) and
recombination (mixing chromosomes) occurduring reproduction
8/6/2019 03 Operations Research
44/66
Use
GA are very well suited when the structureof the search space is not well known
In other words, if the model of our systemis rough, we can use GA to find the best
policy
8/6/2019 03 Operations Research
45/66
How it works
A genetic algorithm maintains a populationof candidate solutions for the problem athand, and makes it evolve byiteratively applying a set of stochastic
operators
Th k l f h
8/6/2019 03 Operations Research
46/66
The skeleton of the
algorithm Generate initial population P(0) t:=0
while not converging do evaluate P(t)
P(t) select best individuals from P(t)
P(t) apply reproduction on P(t)
P(t+1)replace individuals (P(t),P(t)
end while
8/6/2019 03 Operations Research
47/66
Components of a GA
Encoding principles (gene, chromosome) Initialization procedure (creation)
Selection of parents (reproduction) Genetic operators (mutation,
recombination)
Evaluation function (environment) Termination condition
R
8/6/2019 03 Operations Research
48/66
Representation
(encoding) Possible individuals encoding
Bit strings (0101 ... 1100)
Real numbers (43.2 -33.1 ... 0.0 89.2)
Permutations of element (E11 E3 E7 ... E1 E15)
Lists of rules (R1 R2 R3 ... R22 R23)
Program elements (genetic programming)
... any data structure ...
8/6/2019 03 Operations Research
49/66
Choosing the encoding
Use a data structure as close as possible tothe natural representation
Write appropriate genetic operators asneeded If possible, ensure that all genotypes
correspond to feasible solutions
If possible, ensure that genetic operatorspreserve feasibility
8/6/2019 03 Operations Research
50/66
Initialization
Start with a random population a previously saved population a set of solutions provided by a human
expert
a set of solutions provided by anotherheuristic
8/6/2019 03 Operations Research
51/66
Selection
Purpose: to focus the search in promisingregions of the space
Inspiration: Darwins survival of the fittest
Trade-off between exploration and
exploitation of the search space
Li R ki
8/6/2019 03 Operations Research
52/66
Linear Ranking
SelectionBased on sorting of individuals by decreasing
fitness. The probability to be extracted for the ith
individual in the ranking is defined as
where b can be interpreted as the expectedsampling rate of the best individual
L l T
8/6/2019 03 Operations Research
53/66
Extracts k individuals from the populationwith uniform probability (without re-insertion) and makes them play atournament,
where the probability for an individual towin is generally proportional to its fitness
Selection pressure is directly proportional tothe number k of participants
Local Tournament
Selection
R bi i
8/6/2019 03 Operations Research
54/66
Recombination
(Crossover)
Enables the evolutionary process to movetoward promising regions of the searchspace
Matches good parents sub-solutions toconstruct better offspring
8/6/2019 03 Operations Research
55/66
Mutation
Purpose: to simulate the effect of errors thathappen with low probability during
duplication
Results: Movement in the search space
Restoration of lost information to thepopulation
E l i (fi
8/6/2019 03 Operations Research
56/66
Evaluation (fitness
function)
Solution is only as good as the evaluation
function; choosing a good one is often thehardest part
Similar-encoded solutions should have asimilar fitness
8/6/2019 03 Operations Research
57/66
Termination condition
Examples:
A pre-determined number of generations or
time has elapsed
A satisfactory solution has been achieved
No improvement in solution quality hastaken place for a pre-determined number ofgenerations
8/6/2019 03 Operations Research
58/66
Acknowledgements
Part of the material extracted fromIntroduction to Genetic Algorithms byAssaf Zaritsky, Ben Gurion University,
8/6/2019 03 Operations Research
59/66
Simulated Annealing
8/6/2019 03 Operations Research
60/66
The origin
It is the oldest metaheuristic
Originated in statistical mechanics(Metropolis Monte Carlo algorithm)
First presented to solve combinatorial
problems by Kirkpatrick et al. 1983
8/6/2019 03 Operations Research
61/66
The idea
It searches in directions which result insolutions that are of worse quality than the
current solution
It allows to escape local minima
The probability of the exploration of these
unpromising directions is decrementedduring the search
8/6/2019 03 Operations Research
62/66
The algorithm s
GenerateInitialSolution()
Temp:=InitialTemp
while not converging loop
sPickAtRandomFromNeighbor(s)
if J(s) < J(s)
ss else
Accept s as new solution with prob p(T,s,s)
Update(T)
end while
8/6/2019 03 Operations Research
63/66
Boltzmann probability
if s is worse than s, then s might still bechosen as the new solution
the probability depends on d=f(s)-f(s) andon temperature T
the higher d the lower the probability
the higher T the higher the probability
8/6/2019 03 Operations Research
64/66
8/6/2019 03 Operations Research
65/66
Pros and Cons
Pros
proven to converge to the optimum
easy to implement Cons
converges in infinite time the cooling process must be slow
8/6/2019 03 Operations Research
66/66
End of Part II