Cognitive population initialization for swarm intelligence and evolutionary computing

Cognitive computing has been commonly used to address different forms of optimization issues. Swarm intelligence (SI) and evolutionary computing (EC) are population-based intelligent stochastic search techniques promoted to search for their food from the intrinsic way of bee swarming and human evolution. Initialization of populations is a critical factor in the Particle swarm optimization (PSO) algorithm that significantly affects diversity and convergence. Quasi-random sequences based on cognitive computing are more helpful in initializing the population than applying the random distribution for initialization to maximize diversity and convergence. The capacity of PSO is expanded to make it suitable for the optimization problem by adding new initialization techniques based on cognitive computing using the sequence of low discrepancies. The employed low discrepancies sequences included WELL named WE-PSO to solve the optimization problems in large-scale search spaces. The proposed approach has been tested on fifteen well-known uni-modal and multi-modal benchmark test problems extensively used in the literature. Also, WE-PSO efficiency has been compared to standard PSO, and two other Sobol-based PSO (SOB-PSO) and Halton-based PSO (HAL-PSO) initialization approach. The results were obtained to validate the efficiency and effectiveness of the proposed approach. Mean fitness values obtained using WE-PSO designate that WE-PSO is better than standard techniques in multi-modal problems. The computational results also show that the proposed technique outperformed and has a higher accuracy rate than the classical approaches. Besides, the proposed work’s result offers a foresight of how the proposed initialization approach has a significant effect on the importance of cost function, convergence, and diversity.