Data Mining based mutation function for engineering problems with mixed continuous-discrete design variables
| Autor: | Martin Huber, Joon Chung, Kamran Behdinan, Daniel Neufeld, Horst Baier |
|---|---|
| Rok vydání: | 2009 |
| Předmět: |
Continuous optimization
Engineering Mathematical optimization Control and Optimization Optimization problem business.industry computer.software_genre Computer Graphics and Computer-Aided Design Computer Science Applications Engineering optimization Conceptual design Control and Systems Engineering Discrete optimization Genetic algorithm Mutation (genetic algorithm) Data mining Engineering design process business computer Software |
| Zdroj: | Structural and Multidisciplinary Optimization. 41:589-604 |
| ISSN: | 1615-1488 1615-147X |
| DOI: | 10.1007/s00158-009-0439-4 |
| Popis: | Genetic algorithms are well established for solving engineering optimization problems having both continuous and discrete design variables. In this paper, a mutation function for discrete design variables based on Data Mining is introduced. The M5P Data Mining algorithm is used to build rules for the prediction of the optimization objectives with respect to the discrete design variables. The most promising combinations of discrete design variables are then selected in the mutation function of the genetic algorithm GAME to create children. This approach results in faster convergence and better results for both single and multi-objective problems when compared with a standard mutation scheme of discrete design variables. The optimization of a vehicle space frame showed that a mutation probability between 40% and 60% for the discrete design variables results in the fastest convergence. A multi-objective aerospace conceptual design example showed a substantial improvement in the number of pareto-optimal solutions found after 100 generations. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full text from SpringerLink