Memetic Algorithm with Constrained Local Search for Large-Scale Global Optimization

Abstract

AbstractNature-inspired algorithms are seen as potential tools to solve large-scale global optimization problems. Memetic algorithms (MAs) are nature-inspired techniques based on evolutionary computation. MAs are considered as modified genetic algorithms integrated with a local search mechanism. Conventional MAs perform well for small dimensions; however, their performance starts declining with the increase in dimensions. It is popularly known as the “curse of dimensionality” problem. In order to solve this problem, MA with constrained local search (MACLS) is proposed for single-objective optimization problems. MACLS restricts the local search to be performed after every generation. Controlled local search enhances the optimization capability of the MA. MACLS has been evaluated with respect to GS-MPSO (the latest modification of MA) and MLCC, EPUS-PSO, JDEdynNP-F, MTS, DewSAcc, DMS-PSO, LSEDA-gl, UEP, ALPSEA, classical DE (differential evolution), and real-coded CHC algorithms that participated in the Congress on Evolutionary Computation 2008 competition. The results establish that MACLS significantly outperforms these algorithms in attaining global optima for unimodal and multimodal single-objective optimization problems for small as well as large dimensions.