A New Perspective of Improving Initialization for Particle Swarm Optimization
Kamrul Hasan, Juhair Islam, Mohammad Ehsan Shahmi Chowdhury · 2022
Particle Swarm Optimization (PSO) is a pretty efficient algorithm to optimize functions. PSO can be used when traditional optimization algorithms, like Gradient Descent, fail. However, initializing the initial population of the PSO algorithm with random initialization does not yield good results. The primary target of this research is to find a unique way of initializing the initial population, so that PSO performs better than the traditional PSO algorithm. This proposal implemented techniques from Genetic Algorithms and used a novel approach to solve the issue of initialization. Researchers have introduced various initialization techniques in recent years to improve the performance of PSO. Low discrepant sequences like Sobol Sequence, Halton Sequence and Faure sequences have been used to initialize the initial population. But limitation with those initialization technique is that their performance degrades as the search space gets larger. In this research we were able to find better initialization methods, as well as a novel approach. Result comparisons on various scales proves the superiority of the proposed initialization method.