Constrained Evolutionary Optimization
Thomas Philip Rúnarsson, Xin Yao · Kluwer Academic Publishers eBooks · 2006
The penalty function method has been used widely in constrained evolutionary optimization (CEO). This chapter provides an in-depth analysis of the penalty function method from the point of view of search landscape transformation. The analysis leads to the insight that applying different penalty function methods in evolutionary optimization is equivalent to using different selection schemes. Based on this insight, two constraint handling techniques, i.e., stochastic ranking and global competitive ranking, are proposed as selection schemes in CEO. Our experimental results have shown that both techniques performed very well on a set of benchmark functions. Further analysis of the two techniques explains why they are effective: they introduce few local optima except for those defined by the objective functions.