Fitness estimations for evolutionary antenna design

Abstract

Evolutionary design is effective for many applications. However, a choice of the fitness function is often intuitive. In this paper, an effective approach for a comparison of fitness function properties, which uses an estimation of fitness function landscape ruggedness, is described. Penalty coefficients define the importance of each parameter for design goal and change the difficulty of the fitness function landscape significantly. A reasonable choice of objectives and penalty coefficients allows us to reduce the computational efforts because of the smoother landscape of the fitness functions. The effectiveness of the approach is illustrated for fitness functions that are used for evolutionary antenna design.