A Study on Efficient Nonlinear Optimization using the Landscape Modality of the Objective Function based on Neighborhood Structure
| Project/Area Number |
26350443
|
|---|
| 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 | Hiroshima Shudo University |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
海生 直人 広島修道大学, 経済科学部, 教授 (80148741)
廣光 清次郎 広島修道大学, 経済科学部, 研究員 (90043827)
|
|---|
| Project Period (FY) |
2014-04-01 – 2019-03-31
|
| Project Status |
Completed (Fiscal Year 2018)
|
| Budget Amount *help |
¥4,940,000 (Direct Cost: ¥3,800,000、Indirect Cost: ¥1,140,000)
Fiscal Year 2018: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
Fiscal Year 2017: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
Fiscal Year 2016: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
Fiscal Year 2015: ¥1,820,000 (Direct Cost: ¥1,400,000、Indirect Cost: ¥420,000)
Fiscal Year 2014: ¥650,000 (Direct Cost: ¥500,000、Indirect Cost: ¥150,000)
|
|---|
| Keywords | 非線形最適化 / 直接探索法 / 関数形状推定 / 差分進化 / 粒子群最適化 / パラメータ学習 / 近接グラフ / 回転不変性 / メタヒューリスティクス / 集団的最適化 / ペナルティ関数法 / 制約付き最適化 / メタヒューリスティックス / 変数依存性 / 大域探索 / 局所探索 / 教師なし学習 / 低精度近似モデル |
|---|
| Outline of Final Research Achievements |
We proposed methods that improved the efficiency and the robustness of population-based optimization algorithms(POAs), such as differential evolution: 1. Methods that control algorithm parameters (1) according to the landscape modality of the objective function which is estimated by using proximity graphs and neighborhood structures, (2) according to the ranking information of search points. 2. Methods that optimize problems being strong correlation between variables (1) according to the oblique coordinate generated by the points, (2) according to new crossover operations. 3. Method that applies to the penalty coefficient method for POAs where a new point is compared with the old point. The equivalent penalty coefficient value (EPC), which makes the two extended objective values of the points the same, is defined and the method controls the penalty coefficient automatically by EPC. Their advantages are shown by solving benchmark problems and comparing them with other methods.
|
| Academic Significance and Societal Importance of the Research Achievements |
進化的アルゴリズムなどの集団による最適化法は，問題の解析的性質や探索空間の幾何的性質を要請しないため広範囲の問題に適応可能であるが，各問題が持つ個別の性質を有効に活用するという点では問題が残っている．本研究では，関数値のみを利用した目的関数の形状推定や探索点の分類に基づいてアルゴリズムパラメータを適応制御することにより，問題が持つ個別の性質に対応した効率的かつ頑健な探索性能をもつ最適化アルゴリズムを提案した．これにより，広範囲の非線形最適化問題を効率的に解くことが可能になり，様々な分野に応用可能となることが期待できる．
|
Report
(6 results)
Research Products
(50 results)
