Framework for Studying Genetic Optiization of Complex Systems.

摘要

Although functions of interest are generally defined either on some subset of the Cartesian product of the reals or some discrete set, genetic optimization only directly addresses functions defined on diadic groups. As a result the success of genetic optimization relative to the original domain is often dependent on the chosen embedding and representation. This paper analyzes ''ga hard and ga easy'' functions, provides a constructive procedure (that does not use Walsh transforms) to generate hard problems, proves that no single representation is optimal for all functions, and introduces two classes of transformations to change selected hard functions into easy ones. Two procedures for representational change are suggested. 45 refs., 1 fig., 8 tabs.