Two-Stage Learning Brain Storm Optimizer

Abstract

Brain storm optimizer (BSO), a new swarm intelligence paradigm inspired from the human brainstorming process, have received a surge of attentions. However, the original BSO easily suffers from the premature convergence due to its ineffective solution generation operation. In this paper, a two-stage learning strategy is proposed to accelerate the efficiency of the solution generation operation in BSO, thereby enhancing the convergence speed as well as the diversity of population. At the first stage, a learning automaton strategy is conducted to select an appropriate learning exemplar to guide the updating of each solution (i.e., idea). This strategy learns from the feedback information from the environment to enhance the exploration and exploitation. At the second stage, a comprehensive learning strategy is used to generate a set of directional learning exemplars, using utilize useful search experiences during the search. The experimental results on a set of CEC2017 benchmarks validate the effectiveness of the proposed strategy. Then, the resultant algorithm called ACLBSO is applied to resolving the quantitative association rule mining problem. Simulation results show that ACLBSO is a satisfactory optimizer to deal with the complex association rule mining problems.