GEATbx: Genetic and Evolutionary Algorithm Toolbox for use with MATLAB
Documentation

Author: Hartmut Pohlheim,
Version 3.50 (July 2004)

The Genetic and Evolutionary Algorithm Toolbox (GEATbx) provides a set of versatile tools for implementing a wide range of genetic and evolutionary algorithm methods. A first overview can be found in Features and Implementation of the GEA Toolbox.

The documentation of the GEA Toolbox contains:

• a Tutorial explaining the usage of the GEATbx, including quick start, how to write own objective functions and many examples of objective functions.
• an Introduction to Evolutionary Algorithms explaining the structure of Evolutionary Algorithms and their operators as implemented in the GEATbx. Can also be used to understand the working of Evolutionary Algorithms.
• a large Reference section containing a in-deep explanation of all the available parameter setting and options of the GEATbx and the documentation of the implemented functions directly created from the source of the m-files.

1 Tutorial

• Tutorial - Index
• 1 Quick Start
• 2 Variable Representation
• 3 Writing Objective Functions
• 4 Calling Tree of Functions
• 5 Naming Conventions
• 6 Data Structures
• Examples of objective functions
• 1 Parametric Optimization
• 2 Optimization of dynamic systems

2 Introduction to Evolutionary Algorithms

• Evolutionary Algorithms: Principles, Methods and Algorithms
• 1 Introduction
• 2 Overview
• 3 Selection
• 4 Recombination
• 5 Mutation
• 6 Reinsertion
• 7 Parallel implementations
• 8 References (papers, books, conferences) in the area of Evolutionary Computation

3 Reference

• Parameter Settings / Options of GEATbx (This is the place to look for all the features of the GEATbx version 3.x and how to use them.) - in deep overview of features, operators, parameters and options of the toolbox, all settings are shown, the meaning is explained in short
• Index of GEATbx Matlab-Files - documentation, directly produced from the source code of the m-files (purpose, syntax and examples of all routines)

Features

• real, integer, binary (linear and logarithmic scaling, gray coding) and permutation variable representation
• high level functions to all operators
• broad class of operators
• fitness assignement: linear/non-linear ranking, multi-objective ranking (Pareto ranking and goal attainment) and sharing
• selection: stochastic universal sampling, local, truncation, tournament selection
• recombination: discrete, intermediate, line, extended line, permutation/scheduling
• crossover: single/double point, shuffle, reduced surrogate
• mutation: binary, integer, real valued, permutation/scheduling
• reinsertion: global, regional, local
• migration: unrestricted, ring, neighbourhood
• competition: between subpopulations/strategies
• different population models supported (global, regional and local model)
• multiple population support
• multiple strategy support (run multiple search strategies beside each other in one optimization run)
• competition between subpopulations possible (provides efficient distribution of computer resources between different subpopulation)
• sophisticated visualization of state and course of the Evolutionary Algorithms, online (during optimization) and off-line (after an optimization run using logged data)
• comfortable monitoring and storing of results (every run can be fully documented in text and data files)
• incorporation of problem specific knowledge (examples included)

Implementation

• m-file implementation
• compatible on all computer platforms
• modular, user-friendly structure
• high entry functions
• (version 1.9x and 2.x) compatible with Optimization toolbox
• many example and test functions included (ready to run)
• extensive documentation
• tutorial
• quick start
• overview of structure
• explanation of algorithms
• reference of all operators and functions
• written in HTML-format and PDF-format
• default parameter settings
• easily extensible

Installation

Extract the compressed files into the desired directories. The directory structure should be kept intact. It is recommended that the files for the Genetic and Evolutionary Algorithm Toolbox are stored in a directory named geatbx off the main matlab/toolbox directory.

All the paths of the GEA toolbox must be included in the Matlab search path. Really, include all paths.

Restart Matlab and the functionality of the GEATbx should be available. Test it by running one of the demo/example scripts (demo*.m in subdirectory scripts).

Use of documentation

Please consider the following: The use of this documentation is allowed only for personal information. Additionally, you can copy this documentation in unchanged form on your internal net for internal use as documentation of the Genetic and Evolutionary Algorithm Toolbox. It is prohibited to use the documentation or part of it in changed form. (However, if you want to use parts of it, text or graphics for lectures, another documentation or anything else, please contact the author.)

Why a toolbox for Evolutionary Algorithms for Matlab?

During the last years the interest in Genetic and Evolutionary Algorithms (EA: Evolutionary Algorithms) raised steadily. Compared to traditional search and optimization procedures, such as calculus-based and enumerative strategies, Evolutionary Algorithms are robust, globally oriented and generally more straightforward to apply in situations where there is little or no a priori knowledge about the problem to solve. As Evolutionary Algorithms require no derivative information or formal initial estimates of the solution, and because they are stochastic in nature, Evolutionary Algorithms are capable of searching the solution space with more likelihood of finding the global optimum.

Matlab has become a de-facto standard in Computer Aided Control System Design (CACSD). Many areas are catered for by a wide range of toolboxes, notably the Control System, Neural Network and Optimization Toolboxes, and the Simulink/Stateflow simulation package along with extensive visualization and analysis tools. In addition, Matlab has an open and extensible architecture allowing individual users to develop further routines for their own applications. These qualities provide a uniform and familiar environment on which to build genetic and evolutionary algorithm tools.