Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
03/02/2017
1
(ITNPDB8) Evolutionary and Heuristic Optimisation Lecture 4: Population-based Algorithms
Gabriela Ochoa http://www.cs.stir.ac.uk/~goc/
Computing Science and Mathematics,
School of Natural Sciences
University of Stirling, Stirling, Scotland
Outline Optimisation Problems
– The travelling salesman problem – Vehicle routing – Other ‘fun’ and practical examples
Optimisation methods – Heuristics, metaheuristcis – Single point algorithms – Population-based algorithms
• Inspired by evolution by natural selection • Outline of an evolutionary algorithm • Origins of evolutionary algorithms • Main variants (GA, Memetic, ES, GP) • Case studies with DEMOS! (Plants and Creatures) • Other population based approaches
Gabriela Ochoa, [email protected] 2
03/02/2017
2
Travelling salesman problem (TSP)
• A salesman must visit number of cities minimising the total cost of traveling
• Most prominent classical example of an NP-complete combinatorial optimisation problem
• Configurations: permutation (ordering) of cities. – s1= (A B C D), f(s1)= 20+30+12+35= 97
– s2= (A B D C), f(s2)= 20+34+12+42=108
– s3= (A C B D), f(s3)= 42+30+34+35= 141
• Local change (move): swap two cities – s1 = (A B C D)
– s1’ = (A D C B)
Check: Permutation Generator
Gabriela Ochoa, [email protected] 4
03/02/2017
3
The Vehicle Routing Problem
Gabriela Ochoa, [email protected]
• Given: A set of customers, a set of vehicles and a central depot
• Goal: Design minimum cost routes visiting all customers
• Representation: A set of routes one for each truck
• Fitness Function: Reduce the number of trucks used and the total distance travelled
Classification of optimisation algorithms
Gabriela Ochoa, [email protected] 6
• Single solution algorithms are exploitation oriented • Popuplation-based algorithms are exploration oriented
03/02/2017
4
Classification of metaheuristics
Gabriela Ochoa, [email protected] 7
What is an Evolutionary Algorithm?
• EAs fall into the category of “generate and test” algorithms
• They are stochastic, population-based algorithms
• Inspired by the process of Evolution by Natural Selection
• Variation operators (recombination and mutation) create the necessary diversity and thereby facilitate novelty
• Selection reduces diversity and acts as a force pushing quality
03/02/2017
5
Natural Evolution: Fact and Theory
• Changes across successive generations in the heritable characteristics of biological populations
• Theory of Evolution: One of the great intellectual revolutions of human history
• Life on Earth evolved from a universal common ancestor approximately 3.8 billion years ago
Gabriela Ochoa, [email protected]
The Great Tree of Life
Examples of Apparent Design in Nature
Gabriela Ochoa, [email protected]
03/02/2017
6
Evolution by Natural Selection
Natural Selection
1.Variation
2.Hereditary transmission
3.High rate of population growth
4.Differential survival and reproduction
Gabriela Ochoa, [email protected]
Charles Darwin and Alfred Wallace: Theory of evolution by means of
Natural Selection.
1859: On the Origin of Species by Means of Natural Selection: Or, The
Preservation of Favoured Races in the Struggle for Life
Evolutionary Computing: the Analogy
Gabriela Ochoa, [email protected]
Nature Computer
Individual
Population
Fitness
Chromosome
Gene
Crossover and
Mutation
Natural Selection
Solution to a problem
Set of solutions
Quality of a solution
Encoding for a solution
Part of the encoding of a
solution
Search operators
Reuse of good (sub-)
solutions
03/02/2017
7
Outline of an Evolutionary Algorithm
Generate [P(0)]
t 0
WHILE NOT Termination_Criterion [P(t)] DO
Evaluate [P(t)]
P' (t) Select [P(t)]
P''(t) Apply_Variation_Operators [P'(t)]
P(t+1) Replace [P(t), P''(t)]
t t + 1
END
RETURN Best_Solution
Select Parents Select Parents
Recombination Recombination
Mutation Mutation Fitness
Evaluation Fitness
Evaluation
Select Survivors Select Survivors
Initial Population
Important Components
• Representation: Genetic material
• Fitness Function: Task to perform
Origins of evolutionary algorithms
• Evolutionary Programming
– Fogel, Owens, Walsh (1962)
• Evolution Strategy:
– 60s and 70s. I. Rechenberg & H-P Schwefel
• Genetic Algorithms:
– John Holland (1975).
– David Goldberg (1989)
Alan Turing (1912 – 1954). Mathematician, wartime code-breaker and pioneer of
computer science Article: ‘‘Computing Machinery and Intelligence,’’ (1950)
described how evolution and natural selection might be used to automatically
create an intelligent computer program
Google Scholar citations: 63,968
Gabriela Ochoa, [email protected] 14
03/02/2017
8
Genetic algorithms Procedure GA
Generate [P(0)] t = 0 while NOT Termination_Criterion {
Evaluate [P(t)] P' (t) = Select [P(t)] P''(t) = Apply_Operators [P'(t)] P(t+1) = Replace [P(t), P''(t)] t = t + 1
}
Tournament selection
Parent selection: Better individuals get higher chance (proportional to fitness). • Proportional selection (roulette wheel,
stochastic universal sampling) • Scaling methods • Rank selection • Tournament selection • (μ + λ)- and (μ , λ) selection
Replacement (population models) • Generational: each generation set of
parents replaced by the offspring • Steady-state: one offspring is generated
per generation. One member is replaced • Generation gap: a proportion of the
population is replaced
Gabriela Ochoa, [email protected] 15
Search operators for binary representation
Recombination:
• One-point
• N-point
• Uniform
• Pc typically in range (0.6, 0.9)
Mutation:
• Alter each gene independently with
a probability Pm (mutation rate)
• Typically: 1/chromosome_length
Gabriela Ochoa, [email protected] 16
03/02/2017
9
Search operators for permutation representation
Recombination: Combining two permutations into two new permutations:
• choose random crossover point
• copy first parts into children
• create second part by inserting values from other parent:
• in the order they appear there
• beginning after crossover point
• skipping values already in child
8 7 6 4 2 5 3 1
1 3 5 2 4 6 7 8
8 7 6 4 5 1 2 3
1 3 5 6 2 8 7 4
Mutation: Small variation in one permutation, e.g.: swapping values of
two randomly chosen positions,
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
Gabriela Ochoa, [email protected] 17
Memetic (hybrid) algorithms
• Combination of GAs with local search operators, or GAs that use instance specific knowledge in operators
• Orders of magnitude faster and more accurate than GAs, and are the “state-of-the-art” on many problems
(Eiben, Smith, 2003)
• The term meme was coined by R. Dawkins (1976) • The term memetic algorithms by P. Moscato (1989) • The idea of hybridisation in GAs is older
Gabriela Ochoa, [email protected] 18
03/02/2017
10
Other evolutionary algorithms
Evolution Strategies
• Specialised in continuous search spaces: min. f : Rn R
• Rechenberg & Schwefel in the 60s, Technical University of Berlin. Applied to hydrodynamic shape optimisation
• Special feature: self-adaptation of mutation parameters
Gabriela Ochoa, [email protected] 19
Genetic Programming • Evolve a population of computer programs
• Applied to: machine learning tasks (prediction, classification…)
• Representation – Non-linear genomes: trees, graphs
– Linear genomes: grammatical evolution
Evolving Plant-Like Structures
Representation: Lidenmayer Systems (L-systems) • Hungarian Botanist Aristid Lindenmayer (1968)
• A model of morphogenesis, based on formal languages (set of rules and symbols)
Fitness function: What shaped the evolution of plants in Nature • Plants with branching patterns that gather the most light will be
more successful (photosynthesis)
• Need to support vertical branching patterns
03/02/2017
11
L-systems
• Alphabet = {a,b}
• Rules = {a → ab, b → a}
• Axiom: b
b
|
a
└
a b
┘ │
a b a
┘ │ └
a b a a b
_/ / ┘ └ \
a b a a b a b a
Example of a derivation in a L-
System Self-similarity in Nature Computer Graphics
w: F
p: F → F[-F]F[+F][F]
Angle (δ) = 60º
Graphical Interpretation of Strings Brackets [ ]: represent branches
Turtle Graphics (x, y, α) F Move forward a step of length d. (x',y’,α), x'= x
+ d cos(α) and y'= y + d sin(α). Draw segment
+ Turn left by angle δ. The next state of the turtle
is (x,y, α+δ)
- Turn right by angle δ. The next state of the turtle
is (x,y, α-δ)
F
+ -
03/02/2017
12
Simulated Evolution of Virtual Block Creatures
• Virtual creatures (Karl Sims, 1994) • Representation: coded instructions
for their growth and locomotion • Fitness Function: ability to perform a
given task Swimming
Competing Hopping Following
Video
Evolved Antennas for Deployment on NASA’s
Space Technology 5 Mission
Evolved Antennas for Deployment on NASA’s
Space Technology 5 Mission
GECCO-2004, June 2004
J. D. Lohn, G. S. Hornby, D. S. Linden, Evolvable Systems Group, Computational Sciences Division
• Winner of the 2004 Hummies award • Evolved antenna scheduled to fly in space • One of the top evolvable hardware results to date
Previous human designed antenna (helical). Did not meet the requirements.
Evolved Antennas: Better than conventional
design.
03/02/2017
13
Original ST5 Antenna Requirements • Transmit: 8470 MHz
• Receive: 7209.125 MHz
• Gain: >= 0dBic, 40 to 80 degrees
>= 2dBic, 80 degrees
>= 4dBic, 90 degrees
• 50 Ohm impedance
• Voltage Standing Wave Ratio (VSWR): < 1.2 at Transmit Freq
< 1.5 at Receive Freq
• Fit inside a 6” cylinder
• Transmit Frequency: 8470 MHz
• Receive Frequency: 7209.125 MHz
• Antenna RF Input: 1.5W = 1.76 dBW = 31.76 dBm
• VSWR: < 1.2 : 1 at the antenna input port at Transmit Freq, < 1.5 : 1 at the antenna input port at Receive Freq
• Antenna Gain Pattern: Shall be 0 dBic or greater for angles 40 <= theta <=80; 0 <= phi <= 360
• Antenna pattern gain (this shall be obtained with the antenna mounted on the ST5 mock-up) shall be 0.0 dBic (relative to anisotropic circularly polarized reference) for angles 40 <= theta <=80; 0 <= phi <= 360, where theta and phi are the standard spherical coordinate angles as defined in the IEEE Standard Test Procedures for Antennas, with theta=0 to direction perpendicular to the spacecraft top deck. The antenna gain shall be measured in reference to a right hand circular polarized sense (TBR).
• Desired: 0 dBic for theta = 40, 2 dBic for theta = 80, 4 dBic for theta = 90, for 0 <= phi <= 360
• Antenna Input Impedance: 50 ohms at the antenna input port
• Magnetic dipole moment: < 60 mA-cm^2
• Grounding: Cable shields of all coaxial inputs and outputs shall be returned to RF ground at the transponder system chasis. The cases of all comm units will be electrically isolated from the mounting surface to prohibit current flow to the spacecraft baseplate.
• Antenna Size: diameter: < 15.24 cm (6 inches), height: < 15.24 cm (6 inches)
• Antenna Mass: < 165 g.
Parabon Watchman™: Sensor Placement Optimization
Gabriela Ochoa, [email protected]
Optimize sensor placement to cover as much area with as few cameras as possible, in urban environments or rocky terrain.
http://www.parabon.com/ Parabon Computation (Grid Computing)
03/02/2017
14
Other population-based algorithms: the social behaviour metaphor
Ant colony optimisation (ACO)
• Dorigo, Di Caro & Gambardella (1991).
• Inspired by the behaviour of real ant colonies
• A set of software agents artificial ants search for good solutions
• Problem transformed to finding the best path on a weighted graph.
• Ants build solutions incrementally by moving on the graph
• http://www.aco-metaheuristic.org/
• http://www.scholarpedia.org/article/Ant_colony_optimization
Particle Swarm Optimization (PSO)
• Eberhart & Kennedy, 1995
• Inspired by social behaviour of bird flocking or fish schooling
• Solutions (called particles) fly through the search space by following the current optimum particles
• At each iteration they accelerate towards the best locations
• http://www.swarmintelligence.org/
• http://www.scholarpedia.org/article/Particle_swarm_optimization
Gabriela Ochoa, [email protected] 27
Summary
Optimisation Problems – The travelling salesman problem
– Vehicle routing
– Other ‘fun’ and practical examples
Population-based algorithms – Inspired by evolution by natural selection
– Main variants (GA, Hybrid, ES, GP)
– Case studies with DEMOS
– Other population based approaches
Gabriela Ochoa, [email protected] 28