Constraint handling techniques for a non-parametric real-valued estimation distribution algorithm

This article introduces the Non-Parametric Real-valued Estimation Distribution Algorithm (NOPREDA), and its application to constrained optimization problems. NOPREDA approximates the target probability density function by building the cumulative empirical distribution of the decision variables. Relationships and structure among the data is modeled through a rank correlation matrix (Spearmans statistics). The procedure to induce a target rank correlation matrix into the new population is described. NOPREDA is used to solve constrained optimization problems. Three constraint handling techniques are investigated: truncation selection, feasibility tournament, and Stochastic Ranking. NOPREDA´s performance is competitive in problems with inequality constraints. However, a mechanism for properly handling equality constraints remains as part of our future research work.