Fitness estimation models applied to genetic algorithms / Aplicação de modelos de estimação de fitness em algoritmos geneticos
AUTOR(ES)
Francisco Osvaldo Mendes Mota Filho
DATA DE PUBLICAÇÃO
2005
RESUMO
Genetic algorithms usually need a large number of fitness evaluations before a satisfying result can be obtained. In many real-world applications, fitness evaluation may be computationally complex and costly. In these cases, time is an essential subject in performance analysis of genetic algorithms. Therefore, genetic algorithms should provide good solutions in a short period of time. A promising approach to alleviate the computational cost of evaluations considers the fact that sometimes it is better to evaluate only selected individuals and estimate the fitness of the remaining individuals instead of evaluate a whole population. This work suggests the application of fitness estimation models in genetic algorithms. More specifically, it deals with estimation models based on supervised fuzzy clustering (Fuzzy C-Means) and unsupervised fuzzy clustering (Participatory Learning). The goal is to approximate the evaluation functions through the use of fitness estimation models, without significantly affect the quality of solutions. Initially, the fitness estimation models are compared and analyzed experimentally with other models already proposed in the literature. Their performance are evaluated using benchmark optimization problems found in the genetic algorithms literature. Next, the fitness estimation models are used to solve a real-world engineering problem, namely the train scheduling in a freight rail line. This is a typical case where the performance measure of each schedule demands a considerable amount of time. Once again, the performance of the fitness estimation models are evaluated experimentally, comparing their results with the results provided, for simple instances, by linear programming models and, for complex instances, by the classic genetic algorithm
ASSUNTO(S)
teoria da aproximação train scheduling engenharia ferroviaria - planejamento conjuntos difusos approximation models fuzzy clustering algoritmos geneticos genetic algorithms
ACESSO AO ARTIGO
http://libdigi.unicamp.br/document/?code=vtls000377030Documentos Relacionados
- Algoritmos genéticos aplicados à proteção e estimação de harmônicos em sistemas elétricos de potência
- Aplicação de técnicas de planejamento experimental em otimizadores baseados em algoritmos genéticos.
- Função de avaliação dinâmica em algoritmos genéticos aplicados na predição de estruturas tridimensionais de proteínas
- genetic algorithms, power flow convergence
- Seleção de características por meio de algoritmos genéticos para aprimoramento de rankings e de modelos de classificação