Improving a Particle Swarm Optimization Algorithm Using an Evolutionary Algorithm Framework

Particle swarm optimization (PSO) algorithms are now being practiced for more than a decade and have been extended to solve various types of optimization problems. Despite their wide-spread use, in this paper, we investigate standard PSO algorithms for their ability to (i) approach the optimal region and (ii) simultaneously focus in order to find the optimum precisely, in simplistic yet scalable optimization problems exhibiting unimodality. In this study, we observe that a number of commonlyused PSO settings are not able to handle both aspects of optimization as efficiently like that of a previously-known genetic algorithm designed to solve the class of unimodal problems. Here, we highlight the importance of linking different optimization algorithms for their algorithmic equivalence, such that if desired, ideas and operators from one algorithm can be borrowed to enhance the performance of another. We illustrate this aspect by first suggesting an evolutionary algorithm (EA) which is fundamentally equivalent to a PSO and then borrow and introduce different EA-specific operators to the constriction-based PSO. Contrary to the standard PSO algorithms, our final PSO, which performs similar to the existing GA (and outperforms the GA in some occasions), is developed by replacing PSO’s standard child-creation rule with a parent-centric recombination operator. The modified PSO algorithm is also scalable and is found to solve as large as 500-variable problems in a polynomial computational complexity. We emphasize here that EA and related optimization researchers should put more efforts in establishing equivalence between different existing optimization algorithms of interests (such as various genetic and evolutionary and other bio-inspired algorithms) to enhance an algorithm’s performance and also to better understand the scope of operators of different algorithms.