A sinusoidal social learning swarm optimizer for large-scale optimization

Abstract

Large-scale optimization problems are much more difficult compared to traditional optimization problems because they have a larger search space and more numerous local optimum. This paper presents a sinusoidal social learning swarm optimizer (SinSLSO) to effectively tackle large-scale optimization problems. In SinSLSO, sinusoidal function is employed to dynamically adjust the learning probability of particles in the population to balance exploration and exploitation capabilities. Meanwhile, the trapezoidal population size reduction strategy is utilized to make a trade-off between the diversity and convergence speed of SinSLSO. In addition, a new learning strategy is designed to prevent SinSLSO from trapping into a local optimum. Experiments are carried out on two widely used sets of large-scale benchmark functions (i.e., CEC2010 and CEC2013) and the SinSLSO is compared with eleven state-of-the-art algorithms. What is more, the proposed SinSLSO is applied to feature selection problem. The comparison results illustrate the competitive performance of SinSLSO in terms of the quality on most of test problems.