Choosing a starting configuration for particle swarm optimization

Abstract

Abstract-The performance of Particle Swarm Optimization can be improved by strategically selecting the starting positions of the particles. This work suggests the use of generators from centroidal Voronoi tessellations as the starting points for the swarm. The performance of swarms initialized with this method is compared with the standard PSO algorithm on several standard test functions. Results suggest that CVT initialization improves PSO performance in high-dimensional spaces.

A new population initialisation method based on the Pareto 80/20 rule for meta‐heuristic optimisation algorithms

“…Some of the basic initialisation methods include logistic map, LHS, OBL, QBL, and CVT methods, which have also been used in the implementation of the proposed method [1,3,6,7,18,20,26,[48][49][50]. Logistic map creates a chaotic map based on its input parameter.…”

“…For example, the recently proposed algorithms, including Monarch butterfly optimisation [8], slime mould algorithm [9], moth search algorithm [10], supply-demand-based optimisation [11], Harris hawks optimisation [12], mostvaluable-player algorithm [13], multiverse optimiser [14], marine predators algorithm [15], and mayfly algorithm [16], use the uniform random initialisation method to generate their initial population. As the size of the search space increases, the chances of the population covering promising areas of the search space decrease [3,6]. The results of some studies have shown that uniform distributions, especially in high-dimensional space, are not the best initialisation method to solve all problems, and in addition, different initialisation methods lead to different accuracy for different problems [1,3,7,17].…”

“…As the size of the search space increases, the chances of the population covering promising areas of the search space decrease [3,6]. The results of some studies have shown that uniform distributions, especially in high-dimensional space, are not the best initialisation method to solve all problems, and in addition, different initialisation methods lead to different accuracy for different problems [1,3,7,17]. Some initialisation methods use greedy algorithms that use a large population to select the best initial population subset (such as oppositionbased learning [OBL] and quasi-opposition-based learning [QBL] initialisers) [1,18].…”

“…Some initialisers also use exploitation to speed up optimisation processes [5]. Some initialisation methods, such as the centroidal Voronoi tessellation (CVT) method or by biasing the initial population into smaller sizes, can improve the performance of the algorithm for highdimensional space [3,21]. In solving complex problems, combining two or more distribution methods can improve the overall population diversity and, as a result, improve the efficiency and performance of the algorithm [7,20].…”

“…Therefore, initialisation has been considered by researchers with the intention of increasing the diversity of the initial population in order to increase the ability to explore more search space and also to increase the usability while avoiding being trapped in local optimal points by the algorithm. Initialisation for metaheuristic optimisers is usually done randomly and based on the principle of guiding the search space as uniformly as possible [1][2][3][4].…”