Genetic Algorithms for Continuous Problems

Abstract

The single bit mutation and one point crossover operations are most commonly implemented on a chromosome that is encoded as a bit string. If the actual arguments are real numbers this implies a fixed point encoding and decoding each time an argument is updated. A method is presented here for applying these operators to floating point numbers directly, eliminating the need for bit strings The result accurately models the equivalent bit string operations, and is faster overall. Moreover, it provides a better facility for the application of genetic algorithms for continuous optimization problems. As an example, two multimodal functions are used to test the operators, and an adaptive GA in which size and range are varied is tested.