An Analysis of Initialization Techniques of Particle Swarm Optimization Algorithm for Global Optimization

Artificial intelligence (AI) has made its way in almost every field of life. From education and health care, it has its application in expert systems, defense, speech recognition, game development, and many more. Frequently many meta-heuristic techniques have been successfully implied for the detection of medical diseases and promises for accurate perception. A well-known swarm-based intelligent stochastic search technique namely Particle Swarm optimization (PSO) is driven from the key behavior of bee swarm during the process of searching of their food source. Consequently, due to the ease of numerical experimentation, PSO is applied to address various kinds of optimization problems. Population initialization play an important role in PSO algorithm. To improve the convergence and population diversity factors, rather than using the random distribution for initialization of population, quasi random sequences can be used effectively. This study presents a survey of various initialization approaches implied in the PSO that are based on quasi-random sequences (Halton, Torus) and pseudo-random sequence (LCG, MCG) strategies. In this study, the state of the art population initialization techniques are used. The systematic analysis unveils the most potential research areas of population initialization and also existing research gaps, although, the main focus is to provide the directions for future work and development in this particular area. This study gives a systematic view for the state of the-art of research, which is discussed in the specified literature to till date. It is anticipated that this study would be helpful to study the PSO algorithm in more depth and detail. This study also finds the efficiency of many QRS (Halton, Torus) and PRNGs (LCG, MCG) based initialization approaches by comparing their performance observed for nine famous benchmark test problems.