Chapter 3
Ecological Applications of Genetic Aigorithms
D. Morrall
3.1
Introduction
In the early 1960's biologists were attempting to simulate evolution in natural
systems (e.g., Fraser 1960, 1962). About this same time, Holland was working
towards the goal of expressing mathematically the adaptive processes of natural
systems in order to create artificial systems using these processes. Holland' s
1975 publication "Adaptation in natural and artificial systems" provides a
landmark conceptual framework for evolutionary adaptation in artificial systems
using genetic algorithms. His work on genetic algorithms was based on the
premise that natural evolution offers the best model for balancing efficiency and
flexibility in complex systems.
During the 1970's genetic algorithms were primarily the domain of computer
programmers. Programmers studying artificial intelligence techniques were
exploring design attributes of GAs such as mutation and crossover rates, model
behavior, and overall performance.
Goldberg (1989) provides a thorough
overview of the development of genetic algorithms between the 1950's and
1980's. By the 1980's and 90's, genetic algorithms were becoming widely used
as an optimization tool for a variety of real-world applications. Optimization
techniques exploited the standard binary GAs cability to represent and optimize
real-world problems. Some of the most common applications were in the area of
combinatorial optimization. Combinatorial optimization models inc1uded the
c1assical traveling salesman problem (see Goldberg 1989), worker scheduling
(Carnahan et al. 2000), traffic flow (Srinivasan et al. 2000), electrical and waste
routing (Savic and Walters 1997; Song et al. 1997; Rauch and Harremoes 1999;
Su and Lii 1999) and molecular design (Hibbert 1993; Venkatasubramanian et al.
1995). Parametric optimization was somewhat less common than combinatorial
optimization. Examples inc1ude growth media optimization (Weuster-Botz et al.
1995), drug release forrnulation (Hirsch and Muller-Goymann 1995), and
optimization of bioprocess rates (Park et al. 1997).
Classifier systems and control strategies could perhaps be considered the next
generation of GAs. While they are also optimization problems in some sense,
they typically require a more complex structure than the bit-string GA. These
applications often incorporate rules or symbolic capabilities. Pattern recognition
(e.g., Lavine et al. 1999), equation discovery (D' Angelo et al. 1995), consumer
choice (Greene and Smith 1987), fighter plane combat (Smith et al. 2000),
Ecological Applications of Genetic Aigorithms
D. Morrall
3.1
Introduction
In the early 1960's biologists were attempting to simulate evolution in natural
systems (e.g., Fraser 1960, 1962). About this same time, Holland was working
towards the goal of expressing mathematically the adaptive processes of natural
systems in order to create artificial systems using these processes. Holland' s
1975 publication "Adaptation in natural and artificial systems" provides a
landmark conceptual framework for evolutionary adaptation in artificial systems
using genetic algorithms. His work on genetic algorithms was based on the
premise that natural evolution offers the best model for balancing efficiency and
flexibility in complex systems.
During the 1970's genetic algorithms were primarily the domain of computer
programmers. Programmers studying artificial intelligence techniques were
exploring design attributes of GAs such as mutation and crossover rates, model
behavior, and overall performance.
Goldberg (1989) provides a thorough
overview of the development of genetic algorithms between the 1950's and
1980's. By the 1980's and 90's, genetic algorithms were becoming widely used
as an optimization tool for a variety of real-world applications. Optimization
techniques exploited the standard binary GAs cability to represent and optimize
real-world problems. Some of the most common applications were in the area of
combinatorial optimization. Combinatorial optimization models inc1uded the
c1assical traveling salesman problem (see Goldberg 1989), worker scheduling
(Carnahan et al. 2000), traffic flow (Srinivasan et al. 2000), electrical and waste
routing (Savic and Walters 1997; Song et al. 1997; Rauch and Harremoes 1999;
Su and Lii 1999) and molecular design (Hibbert 1993; Venkatasubramanian et al.
1995). Parametric optimization was somewhat less common than combinatorial
optimization. Examples inc1ude growth media optimization (Weuster-Botz et al.
1995), drug release forrnulation (Hirsch and Muller-Goymann 1995), and
optimization of bioprocess rates (Park et al. 1997).
Classifier systems and control strategies could perhaps be considered the next
generation of GAs. While they are also optimization problems in some sense,
they typically require a more complex structure than the bit-string GA. These
applications often incorporate rules or symbolic capabilities. Pattern recognition
(e.g., Lavine et al. 1999), equation discovery (D' Angelo et al. 1995), consumer
choice (Greene and Smith 1987), fighter plane combat (Smith et al. 2000),