An Experimental Comparison of Binary and Floating Point Representations in Genetic Algorithms

Genetic Algorithms (GAs) are innovative search algorithms based on natural phenomena, whose main advantages lie in great robustness and problem independence. So far, GAs were most successful in parameter optimization domains; however, even there certain problems, as lack of fine local tuning capabilities and severe time complexity, prohibit their wider use on most moderately and highly complex problems. Recently, there has been a growing interest in the floating point (FP) representation for genetic algorithms. In this paper we empirically study both FP and binary based GAs using a dynamic control problem—highly complex and quite difficult for any method. Results suggest that the well known advantages of low cardinality alphabets can be compensated for by designing new operators, and that such approach provides means for overcoming some of the mentioned disadvantages.