An Efficient and Exclusively-Feasible Constrained Handling Strategy for Evolutionary Algorithms

Past decades have marked a rising interest in the applicatio n of Evolutionary Algorithms (EAs) for a variety of optimization tasks. Constrained opti mization using EAs is arguably one of the challenging tasks. Out of several existing constrain t handling startegies, we focus on a particular class of methods that are exclusively-feasible i.e. they operate by repairing infeasible solutions and setting them back into the feasible region . In such problems, the success of finding the optima depends upon the choice of (i) the evolutio nary optimizer, (ii) the constraint handling strategy and, (iii) the problem structure. This pa per thoroughly investigates all three aspects in context to constrained optimization using EAs. F ir tly, Particle Swarm Optimization (PSO) algorithm is considered and existing constrained-ha ndling methods are reviewed. The performance of PSO is found to be highly dependent upon the ch oice of the constraint-handling strategy and the location of the optima with respect to the se arch space. In order to overcome the limitations of the current methods two new proposals, na med Inverse Parabolic Methods (IPMs), are made. The existing and proposed constrained han ling methods are also studied with respect to Genetic Algorithms (GAs) and Differential E volution (DE), and importance of the optimiser in order to create efficient solutions is highl i ted. Several simulation results are presented for problems with variable-bounds and applic abi ty of IPMs is also shown for non-linear constraints. A scale-up study upto 500 variable s is also performed and the proposed methods are able to find the optima upto an accuracy of 10 −10, showing the robustness and consistency of the proposed method. The illustrations made in t his paper and guidelines provided along-with should lead to reliable deployment and developm ent of evolutionary methods for constrained-optimisation.