Abstract
Abstract
Particle Swarm Optimization with Migration (MPSO) is proposed to solve the issue that PSO will come across unbearable time cost problem when dealing with High-dimension, Expensive and Black-box objective function tasks. Migration operator is inspired by the migration of Salmon. Salmon will start a dangerous journey from the ocean to the home rivers for reproduction. The process of the entire behavior is similar with the reduction and recovery of dimension. Therefore, we design the Migration operator where a pre-trained Wasserstein Auto-encoders (WAE) is applied to simulates the migration behavior to accelerate the process of evolution in PSO, and we use Least-Squares Regression in lower space to product better generation. In comparison with famous baselines methods in some benchmark functions, MPSO converges more faster and more accurate which show the great potential of migration operation.
Publisher
Research Square Platform LLC