Ensemble of Constraint Handling Techniques for Single Objective Constrained Optimization

Abstract

Many optimization problems in science and engineering involve constraints due to which the feasible region reduces and the search process gets complicated. In addition, when evolutionary algorithms (EAs) are employed to solve constrained optimization problems additional mechanisms referred to as constraint handling techniques are required as EAs generally perform unconstrained search. Generally, the performance of a constraint handling technique depends on its effectiveness in utilizing the information present in the infeasible individuals generated during the evolution process. In the literature, a variety of techniques are developed to exploit the information present in infeasible individuals. However, according to the No Free Lunch (NFL) theorem, no single state-of-the-art constraint handling technique can outperform all others on every problem. In other words, depending on several factors, such as the ratio between feasible search space and the whole search space, multi-modality of the problem, the chosen EA and global exploration/local exploitation stages of the search process, different constraint handling methods can be effective on different problems and during different stages of the search process. Hence, solving a particular constrained problem requires numerous trial-and-error runs to choose a suitable constraint handling technique and to fine-tune the associated parameters. The trial-and-error approach may be unrealistic in applications where the objective function is computationally expensive or solutions are required in real-time.In this chapter, we present an ensemble of constraint handling techniques (ECHT) as an efficient alternative to the trial-and-error-based search for the best constraint handling technique with its best parameters for a given problem. Ensemble being a general concept can be realized with any EA framework. In this chapter, ECHT is combined with an improved differential evolution (DE) algorithm referred to as EPSDE. EPSDE is an improved of DE version based on ensemble framework. The performance of the proposed architecture is compared with the state-of-the-art algorithms.