Design optimization with advanced genetic search strategies

Abstract The present paper describes the capabilities of a modern design optimization tool based on the method of genetic search. This stochastic search technique offers a significantly increased probability of locating the global optimum in a design space with multiple relative optima. The program includes an advanced search technique referred to as directed crossover wherein bit positions on the design strings that offer a higher gain during crossover are assigned higher probabilities of selection as crossover sites. This optimization code also includes a multistage genetic search plan that is useful in problems where the design space is large. Multistage search involves successive refinement in the precision with which design variables are represented in the genetic search process. Also included in this program is a newly developed cluster identification technique that automatically determines the center location and the radius of a hypersphere representing a relative-optimum containing region. Cluster information serves to define accurate parameters required for other advanced techniques such as sharing function implementation and mating restrictions.