Real Coded Genetic Algorithm Operators Embedded in Gravitational Search Algorithm for Continuous Optimization

The objective of this paper is to propose three modified versions of the Gravitational Search Algorithm for continuous optimization problems. Although the Gravitational Search Algorithm is a recently introduced promising memory-less heuristic but its performance is not so satisfactory in multimodal problems particularly during the later iterations. With a view to improve the exploration and exploitation capabilities of GSA, it is hybridized with well-known real coded genetic algorithm operators. The first version is the hybridization of GSA with Laplace Crossover which was initially designed for real coded genetic algorithms. The second version is the hybridization of GSA with Power Mutation which also was initially designed for real coded genetic algorithms. The third version hybridizes the GSA with both the Laplace Crossover and the Power mutation. The performance of the original GSA and the three proposed variants is investigated over a set of 23 benchmark problems considered in the original paper of GSA. Next, all the four variants are implemented on 30 rotated and shifted benchmark problems of CEC 2014. The extensive numerical, graphical and statistical analysis of the results show that the third version incorporating the Laplace Crossover and Power mutation is a definite improvement over the other variants.