A novel adaptive algorithm of particle swarm optimization based on the human social learning intelligence

Abstract

A novel adaptive algorithm of particle swarm optimization is proposed in this paper, where the learning strategy is designed by simulating features of human social learning intelligence. By the principles of this strategy, the swarm is first adaptively grouped into a number of subswarms based on the fitness levels of particles, as well as no need of fixing the population sizes as done in the existing algorithms. Then, the particles in each subswarm constitute three different types classified by their evolutionary levels, which adopt different learning modes defined by distinct adaptive inertia weighting rules and dynamical learning mechanisms. Consequently, all the particles play their respective roles in promoting the global optimization performance of our algorithm, just like all the individuals in society. Numerical tests of our algorithm are conducted by solution of the standard benchmark test problems from the CEC 2014 and CEC 2017 test suites. The test results show the advantages of the proposed learning strategy in this paper, and demonstrate that the overall performance of our algorithm outperforms both of the other compared PSO variants and the other state-of-the-art evolutionary algorithms, in terms of global search ability, solution accuracy and convergence speed. Statistical tests also validate the promising performance of our algorithm.