1-D Parabolic Search Mutation

C. Robertson & R.B. Fisher
Division of Informatics, University of Edinburgh,
Edinburgh, EH1 2QL, UK
craigr@dai.ed.ac.uk

Abstract:

This document describes a new mutation operator for evolutionary algorithms based on a 1-dimensional optimisation strategy. This provides a directed, rather than random, mutation which can increase the speed of convergence when approaching a minimum. We detail typical comparative results of unimodal and polymodal optimisations with and without the operator.

 
• Introduction
  • Canonical EA Definition
  • Canonical EA Operators
  • Building Block Hypothesis and the Problem of Premature Convergence
  • Extensions: Types of Mutation
• Algorithm
  • New Mutation Operator: Parabolic Search Strategy
  • Test Context
• Example Applications
  • Setup
  • Single Minimum Function Optimization
  • Function Optimization with Many Local Minima
    • Sine Product Function
    • Product Function
    • Himmelblau Function
    • Six Hump Camel Function,
    • Generalized Rosenbrock's Function,
    • Quartic Function with Noise,
    • Generalized Rastrigin's Function,
• Results
• Conclusions and Further Work
• References
• About this document ...