Biologically Inspired Non-Mendelian Repair for Constraint Handling in Evolutionary Algorithms

摘要

This paper examines a repair technique that enables evolutionary algorithms to handle constraints. This repair technique, known as GeneRepair, repairs invalid individuals so that all problem constraints are met by every individual in the population. GeneRepair is based on the repair technique used by the Arabidopsis thaliana plant which was proposed by Lolle et al in 2005. This controversial repair method uses information inherited from ancestors previous to the parent (non-Mendelian inheritance) as a repair template to fix errors or invalidities in the current population. We compare the use of three different ancestors as repair templates and investigate the effects of various biological parameters on the choice of repair template to use.