Understanding Crowd Intelligence in Large-scale Systems: A Hierarchical Binary Particle Swarm Optimization Approach

Abstract

As an emerging key technology of crowd intelligence, multi-access edge computing, mobile crowdsensing, and Internet of everything, large-scale optimization can offer suboptimal solutions to the binary optimization problems with NP-complete in these fields. Binary Particle Swarm Optimization (BPSO) is a stable and promising approach with controllable computational complexity. However, it is still challenging to solve these problems by using BPSO. In this paper, inspired by the formulation of crowd intelligence, we propose a hierarchical BPSO algorithm (H-BPSO) based on intelligence model for large-scale binary optimization problems. In H-BPSO, we first formulate the particles in the swarm as entities with intelligence, and divide them into different levels according to their intelligence. Then we design a new strategy for the selection of guiding particles when updating particles. Further, in order to make H-BPSO have better adaptability, and can balance between exploration and exploitation during the evolution, we introduce a dynamic level-number selection strategy. Finally, we investigate the performance of our proposed H-BPSO on a well-known benchmark set of high-dimensional Knapsack instances through comparing H-BPSO with several state-of-the-art BPSO algorithms. The experimental results demonstrate that H-BPSO has better performance when solving high-dimensional Knapsack problems in terms of convergence speed and global search capability.