Transforming the search space with Gray coding

Abstract

Genetic algorithm test functions have typically been designed with properties in numeric space that make it difficult to locate the optimal solution using traditional optimization techniques. The use of Gray coding has been found to enhance the performance of genetic search in some cases. However, Gray coding produces a different function mapping that may have fewer local optima and different relative hyperplane relationships. Therefore, inferences about a function will not necessarily hold when transformed to another search space. In fact, empirical results indicate that some genetic algorithm test functions are significantly altered by Gray coding such that local optimization methods often perform better than genetic algorithms.<>