This paper relies on the idea that a proper search operator, which captures mentioned implicit correlations, can improve performance of evolutionary constrained optimization algorithms. To realize this, an Evolution Strategy (ES) along with a simplified Covariance Matrix Adaptation (CMA) based mutation operator is used with a ranking based constraint-handling method. The proposed algorithm is tested on 13 benchmark problems as well as on a real life design problem. | Yugoslav Journal of Operations Research 24 (2014), Number 3, 307-319 DOI: AN ADAPTIVE ES WITH A RANKING BASED CONSTRAINT HANDLING STRATEGY ALI OSMAN KUSAKCI Faculty of Engineering and Natural Sciences, International University of Sarajevo, Bosnia and Herzegovina akusakci@ MEHMET CAN Faculty of Engineering and Natural Sciences, International University of Sarajevo, Bosnia and Herzegovina mcan@ Received: January 2014 / Accepted: October 2014 Abstract: To solve a constrained optimization problem, equality constraints can be used to eliminate a problem variable. If it is not feasible, the relations imposed implicitly by the constraints can still be exploited. Most conventional constraint handling methods in Evolutionary Algorithms (EAs) do not consider the correlations between problem variables imposed by the constraints. This paper relies on the idea that a proper search operator, which captures mentioned implicit correlations, can improve performance of evolutionary constrained optimization algorithms. To realize this, an Evolution Strategy (ES) along with a simplified Covariance Matrix Adaptation (CMA) based mutation operator is used with a ranking based constraint-handling method. The proposed algorithm is tested on 13 benchmark problems as well as on a real life design problem. The outperformance of the algorithm is significant when compared with conventional ES-based methods. Keywords: Constrained Adaptation. MSC: 65K10, 90C30, 90C59. Optimization, Evolution Strategies, Covariance Matrix 308 Kusakci ., Can M. / An Adaptive ES With a Ranking 1. INTRODUCTION Population based approaches inspired by nature have been widely applied in various scientific domains due to their global search ability and very precise approximation of the global solution [1]. Constrained optimization problems (COPs) of non-convex character requirescomplex optimization methods[2], [3]. While designing new algorithms, the focal point
