Controlling genetic operator rates in evolutionary algorithms
Peter C. Nelson, Fatemeh Vafaee · 2011
The notion of parameterized evolutionary algorithms—EAs in which the behavior of the process is subject to change due to the variation of one or more of the parameters—has been introduced to the field of the Evolutionary Computation from its inception, back to 1970s. It has been realized from the beginning that suitable choices of EA parameters greatly affect the behavior of the algorithm on a particular fitness landscape. Furthermore, an appropriate parameter setting for a given problem domain may be inappropriate for another. Among different parameters, the rates of genetic operators (i.e., mutation and crossover) are known to have a considerable effect on the performance of evolutionary algorithms. Due to the intrinsically dynamic nature of evolutionary processes, the time and the state of the evolution should be inevitably considered in determining the rate of genetic operators. This intuition has led to the emergence of the class of parameter control techniques in which the parameter values are subject to change during an EA run. Motivated by the necessity of controlling genetic operator rates in evolutionary algorithms, the research covered in this work is concerned with introducing a series of new ideas to the field of parameter control techniques. In other words, it aims to take advantage of some new methods, frameworks, or models (e.g., Markov chain model of GA and statistical models of DNA evolution) which to our knowledge are barely used in former parameter control schemes.