Adaptive pattern learning particle swarm optimization for large-scale optimization

초록

Large scale optimization problems (LSOPs) are an important topic in the field of evolutionary computing (EC), and many researchers have designed various learning strategies to try to solve LSOPs more effectively. However, most of the learning strategies are with the fixed learning pattern during the whole evolution process and lack the adaptive adjustment mechanism according to individual property. In fact, different individuals are with different exploitation or exploration abilities, and are suitable for different learning patterns. Therefore, in this paper, we propose adaptive pattern learning particle swarm optimization (APLPSO) to solve LSOPs. In APLPSO, several learning patterns based on different numbers of learning exemplars are first generated to enrich the learning diversity of population. Then, each individual will evaluate the learning patterns and adaptively select its own appropriate learning pattern for updating. The experimental results on two widely used large-scale optimization test suites, CEC2010 and CEC2013, show that APLPSO significantly outperforms other state-of-the-art large-scale optimization algorithms, including the winners of the CEC2010 and CEC2012 competitions. Moreover, we apply APLPSO to a real-world large-scale portfolio optimization application to show its practical applicability. © 2026 Elsevier B.V.

키워드