An evolutionary approach to constrained sampling optimization problems

Constrained sampling optimization problems conform a class of problems where only a part of the solution space is available from any point at any time. That is, one cannot freely choose any point for evaluation at a given time. Evolutionary algorithms are quite inefficient over these problems, as their usual implementations assume that any point in the solution space can be evaluated any time and at no cost. This paper deals with how to modify the general strategy of evolutionary algorithms to address these constraints in an efficient manner and proposes extending their application to other problems that, even though, they are not strictly constrained sampling problems, restricting their sampling capabilities reduces the cost of the optimization procedure without affecting its results. The behavior of the Constrained Sampling Differential Evolution (CS-DE) algorithm is studied as a paradigmatic example of this approach. This study is carried out over a representative set of benchmark functions of different dimensionalities that permit validating the approach and demonstrating its improved efficiency over fitness landscapes with a variety of characteristics.