An alternative constraint handling method for evolution strategies

Abstract

Most real-world search and optimization problems are faced with constraints, which must be satisfied by any acceptable solution. Although a plethora of research is spent on handling constraints in genetic algorithms (GAs), the same is not the case in evolution strategies (ESs). However, this does not say that ESs have not been applied to real-world problems. In fact, in the absence of an efficient constraint-handling technique, ES practitioners have mostly made sure that their ESs started from feasible solutions, a matter which allowed them to apply the commonly-used rejection scheme. In this paper, we borrow a constraint-handling scheme from the GA literature and implement it with standard ES paradigm. The resulting algorithm does not require initial feasible solutions and is found to yield a faster progress in the cylindrical corridor model and be efficient in solving a couple of complicated test problems. The results are interesting and suggest further use of the proposed technique in real-world search and optimization problems.