Optimization approach with particle agents with generic problem-solving capabilities

2023 Fiscal Year Final Research Report

• Project/Area Number: 16K01231
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Social systems engineering/Safety system
• Research Institution: Otaru University of Commerce
• Principal Investigator: Kaji Taichi 小樽商科大学, 商学部, 教授 (60214300)
• Project Period (FY): 2016-04-01 – 2024-03-31
• Keywords: 最適化問題 / メタヒューリスティクス / 粒子群最適化法 / アルゴリズム / 多峰性関数 / 確率的解析 / 極値統計

Outline of Final Research Achievements

Particle Swarm Optimization (PSO), a new paradigm for optimization problems, has been used to obtain highly accurate solutions to multimodal global optimization problems. However, in high-dimensional spaces, PSO has the characteristic of falling into a local solution at an early stage, and its ability is not demonstrated. Therefore, in this study, we investigated a PSO that searches for better quality solutions in higher dimensional spaces without falling into the local solution at an early stage. We improved the ability of PSO by introducing a characteristic probability distribution for the movement of particles and by creating a mechanism for the best construction of the composition of a collection of particles.
In addition, we performed stochastic modeling of the neighborhood that defines the movement of the solution (particles), conducted a mathematical analysis of the movement of particles, and clarified its properties and performance.

Free Research Field

最適化問題

Academic Significance and Societal Importance of the Research Achievements

システムの設計，計画，運用などで最適化手法の利用は欠かせない．しかし，最適化問題は緻密なクラスに分類され，その解法は様々な方法論をとる．提案する高精度なPSOにより汎用的に利用可能な最適化アルゴリズムが存在すれば，あらゆる分野において最適化の活用を推し進める．特に，PSO の欠点である高次元な問題に対して探索力を強化することは，未だ特効薬のない多峰性関数の最適化に対して有効な手立てを提案する特色ある研究となる．
さらに，粒子の近傍の理論的解析を行うことは，PSO のパフォーマンス，限界，およびPSO がもつパラメータによる効果，傾向の特徴が明らかとされる．