A Comprehensive Learning Gravitational Search Algorithm and its Applications

Abstract

Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Diand#64256;erential Evolution (DE), Memetic Algorithm and Gravitational Search Algorithm (GSA) are some eand#64256;ective nature- inspired algorithms with reduced memory requirements computationally eand#64256;ective and more comfortable to implement over computational platforms. This thesis aims to design improvised, more reliable, capable, and eand#64259;cient natural computing methods for various types of optimization problems. The main contribution of the thesis is the proposal of improvised GSA algorithms for global optimization problems. A new comprehensive learning GSA (CLGSA) algorithm is designed for the unconstrained optimization problems. The proposed CLGSA is tested over twenty-eight benchmark problems taken from IEEE CEC sessions 2013. To justify the performance of the CLGSA, it is compared with the eight state-of-the-art algorithms on various measures, including statistical validation, by implementing the statistical test. The eand#64259;ciency and capability of the proposed CLGSA are examined theoretically as well as numerically. The applicability of CLGSA over multimodal problems is investigated in the form of niching CLGSA (nCLGSA), which aims to find all possible global solutions instead of finding one. The performance of nCLGSA is tested over the IEEE CEC-2013 niching benchmark functions which are specially designed for those functions which has more than one global optima. The benchmark contains twelve multimodal problems of different dimensions. The results are compared with the four versions of DE and GSA. For comparison, various measures are taken into account to justify the eand#64259;ciency of the proposed algorithm. Thus, the proposed CLGSA has proven as an efficient algorithm to solve the complex global and multimodal optimization problem in continuous search space. newlineFurthermore, the proposed CLGSA is updated for discrete or binary search space and named as Discrete CLGSA (D-CLGSA). In this, a probability function is used to update the velocity and position in the form of binary d